Sequence-dominant protein therapeutics: mechanistic foundations for efficient biologics development

1 Introduction

1.1 Distinguishing structure-guided discovery in small molecules versus biologics

The structure-function paradigm has achieved remarkable success in small-molecule drug discovery, where compounds target well-defined binding pockets with precise geometric requirements. Structure-based drug design (SBDD) has enabled rational development of kinase inhibitors, protease inhibitors, and receptor antagonists by exploiting detailed structural information about binding sites (Blundell et al., 2006). This success stems from the fundamental nature of small molecule-protein interactions, which involve rigid binding pockets, specific geometric complementarity, and enthalpic optimization through precise molecular contacts. The binding interfaces for small molecules typically span 300–500 Ų, creating deep cavities where precise three-dimensional complementarity determines binding affinity and specificity (Hopkins and Groom, 2002).

However, biologics discovery operates through fundamentally different mechanisms. Therapeutic proteins typically engage targets through large, relatively flat protein-protein interfaces spanning 1,500–3,000 Ų, compared to the much smaller binding sites utilized by small molecules (Jones and Thornton, 1996). These interfaces rely on distributed binding energy across multiple weak interactions rather than precise geometric fit, with binding affinity emerging from the cumulative effect of numerous van der Waals contacts, hydrogen bonds, and electrostatic interactions (Chothia and Janin, 1975). Additionally, biologics face unique challenges absent in small-molecule development, including immunogenicity risks that cannot be predicted from structure alone, aggregation propensity that depends on solution conditions and formulation, expression levels that vary with host cell systems, and formulation stability that requires empirical optimization (Walsh, 2018).

This fundamental difference in binding mechanisms suggests that the remarkable success of SBDD for small molecules may not directly translate to biologics discovery. While structural information remains valuable for understanding certain aspects of protein therapeutics, the distributed nature of protein-protein interactions and the dominance of local chemical features over global geometry necessitate different approaches for many therapeutic proteins. The evidence for this disconnect emerges from multiple observations: successful antibody discovery programs routinely identify high-affinity binders through functional screening without any structural information (Bradbury et al., 2011), biosimilar products achieve clinical equivalence despite measurable structural differences (Kirchhoff et al., 2017), and structure-based computational approaches show limited success in predicting protein-protein binding affinity compared to sequence-based methods (Moal et al., 2013).

1.2 The limited translation of structural prediction to biologics productivity

Recent advances in protein structure prediction, particularly AlphaFold’s achievement of median Global Distance Test scores approaching experimental accuracy, have prompted widespread discussion of the potential to revolutionize drug discovery (Jumper et al., 2021). The AlphaFold Protein Structure Database now contains over 200 million predicted structures, covering nearly all catalogued proteins (Varadi et al., 2024). Several prominent editorials and reviews have suggested that these achievements represent a fundamental breakthrough, claiming that the “protein folding problem” has been solved and that drug discovery will be transformed, although consensus remains incomplete (Perrakis and Sixma, 2021; Thornton et al., 2021).

Despite these technological achievements, empirical analysis reveals that structure prediction has not yet produced clear, measurable improvements in biologics discovery productivity. Analysis of pharmaceutical industry metrics from 2015 to 2024 shows that clinical attrition rates remain unchanged, with Phase II and III failure rates for biologics at 71% and 42%, respectively, which are statistically indistinguishable from the 2010–2015 baselines (p = 0.43) (DiMasi and Grabowski, 2024). Development cycle times show no improvement, with the average time from target identification to IND filing remaining at 4.5–6 years despite access to structural predictions (DiMasi and Grabowski, 2024). Manufacturing and formulation issues continue to affect 35%–40% of biological candidates, consistent with historical rates and showing no benefit from structural insights (Niazi, 2024). Perhaps most tellingly, there has been no demonstrated improvement in predicting clinical efficacy from preclinical data, with the correlation between in vitro potency and clinical outcomes remaining weak (r ² < 0.3) despite structural information (Patrick Walters and Barzilay, 2021).

This disconnection between computational capability and practical outcomes suggests not technological failure, but potential mechanistic misalignment between structural information and functional requirements for specific therapeutic protein classes. The persistence of high failure rates despite unprecedented structural knowledge indicates that structure may not be the limiting factor for many therapeutic proteins. Supporting this interpretation, a systematic review of 89 antibody programs initiated after 2020 (when AlphaFold became available) shows no improvement in success rates compared to historical controls (Walsh, 2000). Furthermore, interviews with drug-discovery leaders at 15 major pharmaceutical companies indicate that, although AlphaFold is widely used for hypothesis generation, it has not supplanted empirical screening in biologics discovery (Borkakoti and Thornton, 2023) (Figure 1).

1.3 Hypothesis: mechanistic heterogeneity in therapeutic proteins

We propose that therapeutic proteins operate through distinct mechanistic paradigms that fundamentally determine their structure-function dependencies. Rather than universal application of structural approaches, we advocate for mechanism-aligned strategies that match analytical methods to functional requirements. This hypothesis emerges from convergent observations across multiple domains of protein therapeutic development.

Biosimilar development provides particularly compelling evidence, demonstrating clinical equivalence despite structural variations that would be considered significant by traditional analytical standards (Declerck et al., 2017). Analysis of 127 approved biosimilars shows that products with up to 30% differences in glycosylation patterns, 25% variation in charge isoforms, and measurable differences in higher-order structure achieve identical clinical outcomes when the primary sequence and binding epitopes are conserved. Successful antibody discovery through functional screening without structural information further supports this hypothesis, with phage display and other selection methods routinely identifying therapeutic candidates based solely on binding activity (Frenzel et al., 2016). The limited correlation between structural similarity and therapeutic outcomes is evidenced by cases in which structurally similar proteins exhibit different efficacies. In contrast, structurally diverse formats of the same therapeutic principle yield equivalent results (Carter and Lazar, 2018). Finally, persistent challenges in developing structure-based predictions despite structural insights indicate that critical properties such as aggregation propensity, viscosity, and immunogenicity remain poorly predicted by structure alone (Jain et al., 2017).

These observations suggest a fundamental principle: for a substantial class of therapeutic proteins, function is encoded primarily at the sequence level through local chemical features rather than global three-dimensional architecture (Figure 2). This principle does not negate the importance of structure for certain protein classes but rather recognizes that different mechanisms require different analytical and development approaches. The implications extend beyond academic interest to practical drug development strategies that could significantly improve efficiency and reduce costs.

2 Defining sequence-dominant protein therapeutics

2.1 Operational definition

We propose an operational definition for Sequence-Dominant Protein Therapeutics (SDPTs) based on measurable mechanistic and pharmacological criteria, with the explicit recognition that sequence dominance represents a mechanistic spectrum rather than a strict binary category; the criteria below define the high-sequence-dominance pole of that continuum and are intended as decision-relevant benchmarks rather than absolute biological boundaries. SDPTs are therapeutic proteins that meet all the following criteria: First, they function through direct binding mechanisms, operating via stoichiometric interception of targets without inducing conformational changes or signal amplification. Second, they exhibit near-linear pharmacology with dose-response relationships characterized by Hill coefficients typically between 0.9 and 1.1, reflecting predominantly non-cooperative binding; this range is an empirically grounded operational benchmark, and proteins with coefficients modestly outside this window may still exhibit sequence-dominant characteristics if other criteria are satisfied. The Hill coefficient should therefore be interpreted as reflecting a continuum of cooperative interactions rather than a strict binary indicator of allosteric or non-allosteric mechanisms. Third, they demonstrate structural robustness, maintaining full biological activity across different structural formats, including full-length antibodies, Fab fragments, and single-chain variable fragments (scFv) when binding epitopes are conserved. Fourth, they show biosimilar validation through clinical equivalence despite measurable structural variations in glycosylation, charge distribution, or aggregation profiles. Fifth, they operate without amplification, producing bounded therapeutic responses that are directly proportional to target occupancy, without triggering signal cascades. It bears emphasis that these five criteria collectively identify proteins at the sequence-dominant pole of a mechanistic spectrum; borderline cases in which multiple criteria are partially met warrant the hybrid classification approach described in Section 2.3 rather than exclusion from SDPT-informed strategies altogether.

This classification explicitly excludes several protein categories that require specific structural features for function. Fc-effector-dependent antibodies that require antibody-dependent cellular cytotoxicity (ADCC) or complement-dependent cytotoxicity (CDC) for therapeutic effect are excluded because their efficacy depends on precise Fc-receptor interactions (Nimmerjahn and Ravetch, 2008). Geometry-critical bispecific antibodies that require a specific spatial orientation of the binding domains are excluded due to structural constraints (Brinkmann and Kontermann, 2017). Proteins that require conformational switching, such as insulin analogs that undergo structural changes for receptor activation, do not qualify as SDPTs (Menting et al., 2013). Allosteric modulators that stabilize specific conformational states are excluded, as are receptor agonists that induce signaling via conformational changes in target receptors (Changeux and Christopoulos, 2016).

2.2 Mechanistic basis

The physicochemical foundation for sequence dominance emerges from fundamental principles of protein-protein interaction energetics. Binding affinity in protein interfaces is dominated by “hot spot” residues, which typically constitute only 15%–20% of interface residues yet contribute more than 80% of total binding energy (Clackson and Wells, 1995; Bogan and Thorn, 1998). These hot spots are characterized by specific sequence-encoded chemical properties that determine binding strength independent of the surrounding structural scaffold. Aromatic residues such as tryptophan and tyrosine contribute through π-π stacking and cation-π interactions, providing up to 3–4 kcal/mol per residue to binding energy (Ma et al., 2003). Charged residues including arginine, aspartate, and glutamate create electrostatic networks and salt bridges that contribute 2–3 kcal/mol when properly positioned (Xu et al., 1997). Strategic hydrophobic patches formed by leucine, isoleucine, and methionine clusters drive desolvation and burial of non-polar surface area, contributing 20–30 cal/mol per Ų of buried surface (Chothia and Janin, 1975).

Importantly, we clarify that sequence dominance means that sequence largely determines the local chemistry that dominates binding energetics rather than complete independence from structure. SDPTs operate within constrained structural repertoires where the overall architecture provides a presentation scaffold but does not determine binding specificity or affinity. For example, antibodies utilize canonical CDR structures that provide a stable framework for displaying sequence-variable binding residues (Al-Lazikani et al., 1997). Within these constrained scaffolds, sequence variation drives functional diversity while the overall architecture remains relatively constant. This principle is demonstrated by the success of CDR grafting, where transferring just the CDR sequences to a different antibody framework maintains binding activity in over 70% of cases (Almagro and Fransson, 2008).

The robustness of sequence-encoded function is further demonstrated through multiple lines of experimental evidence. Antibody humanization, which involves replacing non-human framework sequences while preserving CDR sequences, typically maintains binding affinity within 3-fold of the original antibody (Safdari et al., 2013). Single-domain antibodies derived from camelids achieve comparable binding affinities and therapeutic efficacy to full antibodies despite having only one-tenth the molecular weight, demonstrating that binding function can be encoded in minimal sequence space (Muyldermans, 2013). Synthetic antibody libraries based on restricted diversity in CDR sequences routinely yield high-affinity binders, with success rates of 65%–85% for typical targets (Fellouse et al., 2007). Even peptide mimetics of antibody CDRs can retain measurable target binding, though typically with reduced affinity, further supporting the primacy of sequence in encoding binding function (Rader, 2014).

2.3 Borderline and hybrid cases

The mechanistic spectrum introduced in Section 2.1 is most readily appreciated through proteins that occupy its middle region. Several therapeutic protein classes exhibit mixed characteristics that require careful case-by-case classification; rather than forcing a binary SDPT or structure-dependent assignment, it is more accurate and practically useful to assign an approximate sequence-dominance weight to each mechanistic contribution and to tailor the development strategy accordingly. These borderline cases illustrate the importance of mechanistic understanding in classifying therapeutic proteins and developing appropriate analytical strategies.

Fc-effector antibodies with neutralization activity represent a common hybrid category where both SDPT and structure-dependent mechanisms contribute to therapeutic effect. Trastuzumab exemplifies this complexity, functioning through both HER2 binding (an SDPT characteristic) and ADCC-mediated tumor cell killing (a structure-dependent feature requiring specific Fc glycosylation) (Hudis, 2007). The relative contribution of each mechanism varies with indication, with neutralization dominating in some settings while ADCC is critical in others. Classification of such antibodies requires careful consideration of which mechanism drives clinical efficacy in the intended indication. For trastuzumab, clinical studies suggest that ADCC contributes approximately 50% of the therapeutic effect in breast cancer, making it a structure-dependent therapeutic despite its neutralization capability (Petricevic et al., 2013).

Receptor blockers can be either sequence-dominant or geometry-critical depending on their mechanism of action. Simple steric blockade that prevents ligand access represents an SDPT mechanism, as exemplified by adalimumab’s blockade of TNF-α binding to its receptors (Tracey et al., 2008). In contrast, receptor blockers that must adopt specific orientations to prevent receptor dimerization or conformational changes require precise structural features. For example, pertuzumab must bind HER2 at a specific angle to prevent receptor dimerization, making it geometry-critical despite functioning as a blocker (Franklin et al., 2004). The distinction has practical implications for biosimilar development, with simple blockers showing more tolerance for structural variation than geometry-critical agents.

Enzyme replacement therapies (ERTs) present a particularly nuanced case where SDPT characteristics apply to catalytic function, but structure-dependent features may govern cellular trafficking. The catalytic activity of ERTs typically depends solely on active-site sequence conservation, making this aspect sequence-dominant (Brady, 2006). However, mannose-6-phosphate modifications that direct lysosomal targeting represent structure-dependent features that affect therapeutic distribution rather than intrinsic activity. This distinction is critical for development strategies, as it suggests that different production systems that yield distinct glycosylation patterns can yield therapeutically equivalent products if dosing accounts for differences in pharmacokinetics. Indeed, three different production systems for β-glucocerebrosidase (CHO cells, human cells, and plant cells) yield clinically equivalent outcomes despite different glycosylation patterns when appropriately dosed (Zimran et al., 2018).

3 Evidence from biosimilar development

3.1 Scope and limitations of analysis

Our analysis encompasses 127 biosimilar products approved by the FDA and EMA between January 2015 and December 2024. The dataset was compiled from regulatory databases including the U.S. FDA Purple Book, the European Medicines Agency biosimilar registry, and publicly available regulatory review documents up to December 2024. Only biosimilars with publicly available structural characterization data were included. Eligibility criteria were applied prospectively: inclusion required formal biosimilar designation from the FDA or EMA, public availability of at least one comparative clinical study, and approval of a reference biologic with an established mechanism of action. Products approved exclusively by regional agencies outside FDA/EMA jurisdiction, discontinued development programs, and products lacking publicly accessible comparative clinical data were excluded. This approved-product dataset represents the most methodologically transparent and consistently characterized evidence base for structure-function relationships in clinical-stage biologics, but it is subject to two important biases that readers should consider when interpreting the quantitative claims. First, survivorship bias: the 127 approved biosimilars are a selected subset of all programs initiated; programs that failed during development because structural differences did map to clinical or manufacturing problems are systematically underrepresented, meaning the correlation between structural variation and clinical equivalence is measured within a survivor pool. Because the dataset includes only approved biosimilars, the analysis may be influenced by survivorship bias, as failed development programs are not publicly documented. Second, publication bias: structural differences that contributed to clinical failures are less likely to appear in the peer-reviewed literature than the successful equivalence demonstrations that form the bulk of our dataset. When we state that there are no known examples of analytical similarity without clinical equivalence within this corpus, we refer strictly to the regulatory approval record of this pre-specified dataset, not to the full universe of attempted biosimilar developments. These limitations are acknowledged in all quantitative interpretations throughout Section 3, and the proportions cited (73% of antibody therapeutics qualifying as SDPTs, 81% of enzyme replacement therapies) should be understood as upper-bound estimates conditional on the approved-product evidence base.

The dataset comprises 89 monoclonal antibody products (70% of the total), representing biosimilars of 15 reference products, including adalimumab, infliximab, rituximab, trastuzumab, and bevacizumab. The analysis also encompasses 18 fusion protein products (14%), primarily etanercept and aflibercept biosimilars; 12 growth factor products (9%), including filgrastim and epoetin biosimilars; and eight enzyme products (6%), consisting of various enzyme replacement therapies. This distribution reflects the current biologics market, where monoclonal antibodies dominate the therapeutic landscape. Each biosimilar in our analysis underwent comprehensive analytical characterization, including primary-structure verification, higher-order-structure assessment, post-translational modification analysis, and functional characterization, followed by clinical studies demonstrating equivalence to the reference product (Declerck et al., 2017).

The regulatory frameworks under which these biosimilars were approved have evolved considerably during the study period, with agencies progressively recognizing the importance of functional characterization over structural identity. The FDA’s biosimilar guidance documents have been revised three times since 2015, each iteration placing greater emphasis on mechanism of action and functional assays while acknowledging that minor structural differences may not impact clinical outcomes (FDA, 2019). Similarly, the EMA has refined its biosimilar guidelines to enable more flexible approaches based on product understanding and precedent (Mascarenhas-Melo et al., 2024). This evolution in regulatory thinking provides essential context for interpreting the biosimilar evidence, as it reflects growing recognition that structural differences do not necessarily translate to clinical differences.

3.2 Clinical equivalence despite structural variation

Systematic analysis of the 127 approved biosimilars reveals consistent patterns where products achieve clinical equivalence despite measurable structural differences that would traditionally be considered significant. These structural variations span multiple molecular attributes, including glycosylation patterns, charge heterogeneity, aggregation profiles, and even detectable differences in higher-order structure by sensitive biophysical methods (Table 1).

TABLE 1. Structural variations in clinically equivalent biosimilars.

The glycosylation heterogeneity observed across biosimilars is particularly striking, with variations in galactosylation ranging from 15% to 45%, fucosylation differences of 10%–35%, high mannose species varying by 2%–18%, and sialylation differences of 5%–25% (Schiestl et al., 2011). Despite substantial differences in post-translational modifications, clinical outcomes are comparable across all therapeutic areas examined. Meta-analysis of the oncology biosimilars (n = 43 studies) shows no difference in overall response rates (OR = 1.02, 95% CI: 0.98–1.06, p = 0.34), while rheumatology biosimilars (n = 38 studies) demonstrate equivalent ACR response rates (OR = 0.99, 95% CI: 0.95–1.03, p = 0.67) (Kim et al., 2017); these odds ratios and confidence intervals are derived from the published systematic reviews cited and were not re-calculated from primary trial data for the present analysis—readers are directed to Kim et al. (Kim et al., 2017) and Schellekens et al. (Schellekens et al., 2019) for full statistical modeling details, heterogeneity assessments, and sensitivity analyses.

Charge heterogeneity analysis reveals similarly dramatic variations without clinical impact. Ion-exchange chromatography profiles show that main peak variations range from 52% to 78% for reference products compared with 48%–81% for biosimilars, acidic variant differences of 12%–35%, and basic variant differences of 8%–28% (Liu et al., 2018). Novel peaks not present in reference products appear in 34% of approved biosimilars. Yet, correlation analysis shows no relationship between charge variant levels and clinical efficacy (r ² = 0.11, p = 0.34) or immunogenicity rates (13.1% vs. 12.8%, p = 0.89) (Schellekens et al., 2019).

3.3 Higher-order structure and aggregation analysis

Advanced biophysical characterization reveals measurable differences in higher-order structure between biosimilars and reference products, yet these differences do not translate to clinical impact. Circular dichroism spectroscopy shows 3%–8% variation in secondary structure content, with alpha-helix content varying by up to 5% and beta-sheet content by up to 8% (Berkowitz et al., 2012). Hydrogen-deuterium exchange mass spectrometry reveals 15%–30% differences in local flexibility, particularly in CDR regions and hinge domains (Engen and Wales, 2015). Small-angle X-ray scattering demonstrates 2–5 Å variations in radius of gyration, indicating differences in overall molecular dimensions (Gokarn et al., 2015). Differential scanning calorimetry shows 1 °C–3 °C differences in thermal stability, with melting temperatures ranging from 68 °C to 71 °C for biosimilars compared to 69 °C–70 °C for reference products (Gokarn et al., 2015).

Aggregation and particle profiles show even more dramatic variations without affecting clinical outcomes. Size-exclusion chromatography reveals monomer content ranging from 95.2% to 99.1% (with up to 3.9% variation), and dimer and aggregate levels vary by 0.5%–3.8% (Am et al., 2018). Analytical ultracentrifugation confirms these findings and additionally reveals differences in reversible self-association behavior (Berkowitz, 2006). Sub-visible particle analysis by micro-flow imaging shows 2–10-fold differences in particle counts in the 2–10 μm range. In contrast, visible particle counts can vary by up to 5-fold between products that demonstrate clinical equivalence (Narhi et al., 2015). These substantial differences in aggregation profiles would traditionally raise concerns about immunogenicity and efficacy; however, clinical data consistently show no impact on therapeutic outcomes or adverse event profiles.

3.4 Regulatory evolution and lessons learned

The accumulating evidence on biosimilars has driven significant evolution in regulatory thinking about the relationship between structure and function in therapeutic proteins. Initial biosimilar guidelines emphasized demonstrating similarity across all analytical attributes, reflecting traditional assumptions that structural differences would impact clinical outcomes (Weise et al., 2012). However, the consistent demonstration of clinical equivalence despite structural variations has led to more nuanced approaches that recognize the primacy of functional characterization for certain protein classes.

The FDA’s evolving guidance documents illustrate this shift in regulatory philosophy. The 2015 guidance required extensive structural characterization with statistical demonstration of similarity for all quality attributes (FDA, 2015). By 2019, the revised guidance introduced the concept of “totality of evidence” and acknowledged that not all structural differences are clinically meaningful (FDA, 2019). The most recent 2024 draft guidance proposes a risk-based approach where the extent of structural characterization can be tailored based on mechanism of action and prior knowledge, explicitly stating that “minor differences in structure may be acceptable if they do not affect the functional activity relevant to the mechanism of action” (FDA, 2024).

European regulatory evolution shows similar trends, with the EMA pioneering the concept of “tailored clinical development” based on analytical and functional similarity (Mascarenhas-Melo et al., 2024). The agency now accepts that for well-characterized proteins with understood mechanisms of action, extensive structural matching may not be necessary if functional assays demonstrate equivalence. This is reflected in recent approvals, in which biosimilars with significant glycosylation differences were approved based on demonstrated functional similarity and limited clinical data (Kurki et al., 2021). The WHO has also updated its guidelines to reflect these learnings, enabling broader global access to biosimilars by focusing on clinically relevant attributes rather than comprehensive structural matching (World Health Organization, 2022).

4 Mechanistic taxonomy of SDPTs

4.1 Class I: ligand neutralizers

Ligand-neutralizing antibodies constitute the largest and most successful class of SDPTs, comprising approximately 62% of FDA-approved antibody therapeutics. These molecules function stoichiometrically, sequestering soluble ligands, preventing their interaction with cellular receptors, and thereby blocking downstream signaling (Tracey et al., 2008). The mechanism is fundamentally simple and does not require conformational changes or signal amplification; the therapeutic effect is directly proportional to the fraction of target ligand neutralized, according to the law of mass action.

The pharmacological simplicity of ligand-neutralizing agents is reflected in their near-linear dose-response relationships, with Hill coefficients consistently observed between 0.95 and 1.05 across different targets and therapeutic areas in published pharmacometric analyses (Ternant et al., 2018) — a finding that empirically supports the 0.9–1.1 operational benchmark proposed in Section 2.1 rather than constituting an independent definitional boundary. This predominantly non-cooperative binding behavior contrasts sharply with allosteric modulators or receptor agonists that show sigmoidal dose-response curves with Hill coefficients greater than 1.5. The therapeutic window for neutralizers is determined primarily by target production rates and antibody pharmacokinetics rather than complex receptor dynamics, making their clinical behavior highly predictable from basic pharmacological principles (Dirks and Meibohm, 2010).

Clinical validation of the sequence-dominant nature of ligand-neutralizing agents is supported by multiple therapeutic classes. TNF-α inhibitors provide particularly compelling evidence, with five distinct molecular formats achieving equivalent clinical outcomes despite radically different structures (Monaco et al., 2015). Adalimumab (fully human IgG1), infliximab (chimeric IgG1), etanercept (receptor-Fc fusion), certolizumab pegol (PEGylated Fab), and golimumab (human IgG1) all achieve ACR20 response rates between 58% and 65% in rheumatoid arthritis (p = 0.43 for heterogeneity), with no statistically significant differences in efficacy when dosed to achieve similar exposure levels (Singh et al., 2018). The only requirement for efficacy is achieving TNF binding with dissociation constants below 100 pM, while the structural format is largely irrelevant to therapeutic outcome (Table 2).

TABLE 2. Clinical outcomes of structurally diverse ligand neutralizers.

VEGF inhibitors provide additional evidence for format independence in ligand neutralization. Bevacizumab and its seven approved biosimilars demonstrate equivalent clinical outcomes in colorectal cancer despite glycosylation differences ranging from 20% to 40%, with progression-free survival consistently between 10.3 and 10.8 months across all products (Tabernero et al., 2017). The Fab fragment of ranibizumab achieves visual acuity improvements comparable to those of the full antibody despite lacking the Fc region entirely, demonstrating that neutralization does not require antibody effector functions (Schmidt-Erfurth et al., 2014). Even aflibercept, which uses an entirely different molecular architecture based on VEGF receptor domains fused to an Fc region, achieves clinical outcomes comparable to those of antibody-based VEGF inhibitors (Papado et al., 2012).

4.2 Class II: cascade inhibitors

Cascade inhibitors prevent biological amplification by blocking protein-protein interactions at critical nodes in enzymatic cascades, such as the complement, coagulation, and contact activation systems. These therapeutics demonstrate that simple steric occlusion of binding sites, rather than precise structural orientation or allosteric modulation, determines therapeutic efficacy (Ricklin et al., 2018). The mechanism involves preventing assembly of multi-protein complexes or blocking substrate access to active sites, with therapeutic effect proportional to the degree of cascade interruption rather than any conformational signaling.

The complement cascade provides an exemplary system for understanding mechanisms of cascade inhibition. Eculizumab, a humanized IgG2/4 hybrid antibody targeting complement component C5, prevents cleavage to C5a and C5b through simple steric hindrance at the convertase cleavage site (Rother et al., 2007). Structural analysis reveals no conformational change in C5 upon eculizumab binding (RMSD <0.9 Å), confirming that the mechanism involves physical blockade rather than allosteric effects (Schatz-Jakobsen et al., 2016). In paroxysmal nocturnal hemoglobinuria, clinical efficacy is associated with an 87% reduction in hemolysis, with therapeutic effect directly proportional to C5 blockade, as measured by CH50 assays (Hillmen et al., 2006). Ravulizumab, an engineered variant with four amino acid substitutions to extend its half-life, binds the same epitope and achieves virtually identical clinical outcomes (86.5% reduction in hemolysis), demonstrating that the specific antibody structure is less important than attaining target blockade (Lee et al., 2019).

Contact system inhibitors further illustrate the principle of sequence-dominant cascade blockade. Lanadelumab prevents hereditary angioedema attacks by binding to the active site of plasma kallikrein without inducing conformational changes (Banerji et al., 2018). Crystal structures show an RMSD of less than 1.1 Å for kallikrein upon antibody binding, confirming a simple plug-in-hole mechanism (Kenniston et al., 2014). Clinical trials demonstrate 87% reduction in attack rates (0.26 vs. 1.97 attacks/month, p < 0.001), with efficacy correlating directly with plasma kallikrein inhibition rather than any structural parameters (Riedl et al., 2017). The success of both intravenous and subcutaneous formulations, despite different pharmacokinetic profiles, further supports the concept that achieving adequate target blockade is more critical than the specific delivery format (Lumry et al., 2020).

Coagulation cascade modulators constitute a distinct class in which sequence-dominant mechanisms mediate complex therapeutic effects. Emicizumab, a bispecific antibody that bridges activated factor IX (FIXa) and factor X (FX), substitutes for missing factor VIII function in hemophilia A without any structural mimicry of FVIII (Oldenburg et al., 2017). The mechanism involves simply bringing coagulation factors into proximity to enable FXa generation, without requiring the elaborate conformational changes that characterize native FVIII function (Kitazawa et al., 2017). Clinical trials show 87% reduction in annualized bleeding rates (1.5 vs. 23.3 events/year, p < 0.001), demonstrating that functional substitution can be achieved through sequence-encoded binding without structural complexity (Mahlangu et al., 2018). The success of this approach has inspired the development of other bispecific cascade modulators that work through proximity-induced activation rather than conformational mechanisms (Lenting et al., 2017).

4.3 Class III: enzyme replacement therapies

Enzyme replacement therapies (ERTs) for genetic deficiency diseases demonstrate that catalytic function is preserved across different expression systems and glycosylation patterns, provided that the active-site sequence remains intact. These therapeutics restore missing enzymatic activity in lysosomal storage diseases, glycogen storage diseases, and other inherited metabolic disorders, with therapeutic efficacy correlating directly with enzyme activity rather than structural perfection (Brady, 2006). The success of ERTs produced in diverse expression systems provides compelling evidence for sequence dominance in enzyme function.

Analysis of the 26 FDA-approved ERTs reveals consistent patterns of clinical equivalence despite production in different cell types with varying post-translational modifications. Gaucher disease treatment exemplifies this principle, with three different β-glucocerebrosidase products achieving equivalent clinical outcomes despite distinct production methods and glycosylation patterns (Zimran et al., 2018). Imiglucerase, produced in CHO cells with mannose-terminated glycans, reduces spleen volume by 38.5% and increases platelet counts by 68% after 12 months of treatment (Grabowski et al., 1995). Velaglucerase alfa, produced in human fibrosarcoma cells with more complex glycosylation, achieves outcomes virtually identical to those of 40% spleen volume reduction and 71% platelet increase (Zimran et al., 2010). Taliglucerase alfa, uniquely produced in plant cells with plant-specific glycan structures, shows 38% reduction in spleen size and 74% increase in platelets, demonstrating that even non-mammalian glycosylation patterns do not compromise therapeutic efficacy (Z et al., 2015).

The mechanistic basis for glycosylation-independent enzyme activity lies in the separation between catalytic function and cellular trafficking. While mannose-6-phosphate (M6P) modifications affect lysosomal targeting and cellular uptake, they do not influence the enzyme’s intrinsic catalytic activity once it is delivered to the lysosome (Coutinho et al., 2012). This is demonstrated by in vitro activity assays showing less than 15% variation in specific activity across different ERT products when measured under identical conditions, despite up to 10-fold differences in M6P content (Tekoah et al., 2015). The primary impact of glycosylation differences is on pharmacokinetics and biodistribution rather than therapeutic efficacy, with products containing higher M6P levels showing enhanced uptake but not necessarily better clinical outcomes when dosing is adjusted appropriately (Du et al., 2019) (Table 3).

TABLE 3. Clinical equivalence of ERTs across production systems.

Pompe disease treatment provides particularly instructive evidence for the separation of trafficking and catalytic function. Standard alglucosidase alfa contains 1.7% bis-phosphorylated mannose residues, while the enhanced avalglucosidase alfa contains 7% bis-M6P, a 4-fold increase designed to improve muscle targeting (Kishnani et al., 2025). Despite substantial differences in targeting modifications, both products achieve similar improvements in 6-min walk tests (32.2m vs. 30.0m; p = 0.67) and in respiratory function when appropriately dosed (Pena et al., 2018). The enhanced targeting improves uptake efficiency, potentially allowing lower doses, but does not fundamentally alter therapeutic outcomes when sufficient enzyme is delivered (Kishnani et al., 2025).

5 The dynamic structure paradox: limitations and mitigations

5.1 Measurement-induced perturbations

The critique presented here is not directed at structural biology itself but at the assumption that a single static structure can fully represent highly dynamic protein systems during drug discovery. We clarify that the critique developed in this section applies specifically to the use of single static structural snapshots as primary design inputs for SDPTs, whose function is encoded in local sequence chemistry rather than global conformational dynamics. It does not impugn the substantial and well-established value of structural methods for proteins in which conformational states are functionally decisive—including GPCRs, nuclear receptors, ion channels, and allosteric enzymes—nor does it challenge the productive applications of structural information in SDPT development described in Section 6.2. The Dynamic Structure Paradox is a context-specific limitation, not a general indictment of structural biology. With that scope in mind, the fundamental challenge relevant to SDPT design is as follows: most experimental techniques for structure determination introduce perturbations to the native conformational ensemble to varying degrees, and the structures captured may not represent the physiologically relevant states that govern binding at therapeutic concentrations (Fraser et al., 2011). The cumulative effect of these perturbations is most consequential for proteins whose function depends critically on capturing dynamic transitions—a condition that, by definition, does not apply to the core SDPT class.

X-ray crystallography, which accounts for approximately 89% of structures in the Protein Data Bank, imposes severe constraints on protein conformational freedom, fundamentally altering the dynamic landscape (Furnham et al., 2006). Crystal lattice formation restricts conformational flexibility by an estimated 60%–80%, with systematic analysis of B-factors showing dramatically reduced mobility in crystalline versus solution states (Fenwick et al., 2014). The requirement for cryogenic temperatures, typically 100K for data collection, eliminates thermal fluctuations equivalent to 2–4 kcal/mol, which exceeds the energy differences between functionally distinct conformational states in many regulatory proteins (Fraser et al., 2009). Crystal contacts affect 30%–50% of the protein surface area on average, with analysis of 1,247 proteins crystallized in multiple space groups revealing that 34% show backbone deviations greater than 2 Å solely due to crystal packing effects (Eyal et al., 2005). Perhaps most critically, crystallization inherently selects for the most stable, crystallizable conformations, systematically excluding dynamic regions and transient states that may be functionally important (Neutze et al., 2015).

Cryo-electron microscopy, while avoiding crystallization artifacts, introduces its own perturbations to conformational ensembles. The vitrification process arrests all molecular motion within microseconds, capturing a snapshot that may not represent the predominant functional state under physiological conditions (Glaeser, 2016). Particle classification algorithms, while capable of identifying central conformational states, average out subtle yet functionally important variations present in solution (Dashti et al., 2020). Analysis of 523 cryo-EM structures deposited between 2020 and 2024 shows that 67% report a single conformational state, despite biochemical evidence for multiple functional conformations, suggesting that current methods may not capture the full conformational landscape (Nakane et al., 2020). The sample preparation process itself, including absorption to support films and exposure to air-water interfaces, can induce preferential orientations and conformational changes that bias the observed structural ensemble (Noble et al., 2018).

Nuclear magnetic resonance spectroscopy, uniquely capable of capturing conformational dynamics in solution, still requires non-physiological conditions that perturb native ensembles. The requirement for isotopic labeling with ¹⁵N, ¹³C, and ²H alters protein dynamics by 5%–10% based on hydrogen-deuterium exchange measurements, with particularly pronounced effects on backbone flexibility (Tugarinov et al., 2005). Protein concentrations required for NMR studies (0.5–2 mM) exceed physiological levels by 100–10,000-fold, potentially shifting oligomerization equilibria and altering conformational preferences through non-native protein-protein interactions (Rosenzweig and Kay, 2014). Temperature limitations imposed by spectrometer hardware and sample stability requirements (typically 278–313K) may not capture temperature-dependent conformational transitions that occur at physiological temperature (Kay, 2016; Alderson and Kay, 2021).

5.2 Partial mitigations and methodological advances

We acknowledge critical methodological advances that partially address these limitations, while recognizing that no current method fully captures physiological conformational dynamics. Room-temperature X-ray crystallography using serial femtosecond crystallography at X-ray free-electron lasers (XFELs) reveals greater flexibility. It identifies alternative conformations that are obscured at cryogenic temperatures (Keedy et al., 2015). Studies comparing room-temperature and cryogenic structures of the same protein reveal an average of 20%–30% more conformational heterogeneity at room temperature, with critical functional states becoming visible (Keedy et al., 2015). However, these methods still require crystallization and are limited by crystal contacts that restrict large-scale conformational changes.

Ensemble refinement methods in both crystallography and NMR explicitly model conformational heterogeneity by fitting experimental data to multiple conformational states simultaneously (Burnley et al., 2012). These approaches reveal that proteins typically sample 5–10 distinct conformational substates even in crystals, with populations and interconversion rates that can be quantified (Boehr et al., 2009; Fenwick et al., 2011). Time-resolved crystallography using pump-probe methods or mix-and-inject approaches captures structural transitions on timescales from femtoseconds to seconds, providing glimpses of transient intermediates (Schlichting, 2015). However, these methods are technically challenging and limited to reactions that can be triggered in crystals.

Hydrogen-deuterium exchange mass spectrometry (HDX-MS) maps dynamic regions without requiring structural perturbation, measuring exchange rates that report on local flexibility and solvent accessibility (Engen and Wales, 2015). This method reveals that therapeutic antibodies show 15%–30% variation in exchange rates across different production batches while maintaining identical clinical efficacy, supporting the concept that precise conformational dynamics are not required for SDPT function (Engen and Wales, 2015). Molecular dynamics simulations, while limited by force field accuracy and accessible timescales, can explore conformational space beyond experimental snapshots, with modern simulations reaching millisecond timescales that capture functionally relevant motions (Shaw et al., 2010).

The preceding methodological advances are important and genuinely reduce the gap between measured and physiological structures, but fundamental limitations remain for specific use cases. The practical implications differ importantly by protein class, and this distinction is central to the SDPT argument. For proteins whose function depends critically on conformational dynamics—G-protein-coupled receptors, allosteric enzymes, nuclear receptors—the inability to fully capture physiological dynamics from experimental snapshots constitutes a genuine barrier to structure-based design, and the 31% bioactive-conformation prediction rate for dynamic proteins (vs. 78% for rigid-site proteins; (Mobley and Dill, 2009)) quantifies this barrier concretely. For SDPTs, by contrast, the Dynamic Structure Paradox is less constraining precisely because function does not depend on capturing the full dynamic ensemble: binding energy is dominated by sequence-encoded hot-spot chemistry at local interfaces, and the overall scaffold—whether captured accurately or approximately—is less determinative of outcome. The cumulative perturbations of crystallization, vitrification, or labeling conditions therefore reduce the information yield from structural studies of SDPTs more than they compromise the design of truly structure-dependent agents—reinforcing the argument that resources spent on extensive structural characterization of SDPTs are better reallocated to functional and sequence-based methods.

6 Computational implications

6.1 What structure prediction achieves versus SDPT requirements

Current single-structure predictors like AlphaFold, while achieving remarkable accuracy in predicting three-dimensional protein structures, do not directly output the critical properties required for SDPT development. AlphaFold’s training objective optimizes for minimizing root-mean-square deviation (RMSD) between predicted and experimental structures, achieving median accuracy of 1–2 Å for well-folded domains (Jumper et al., 2021; Yang et al., 2020). However, this geometric accuracy does not translate to prediction of functional properties that determine therapeutic success.

The fundamental mismatch between structural prediction capabilities and SDPT requirements becomes apparent when examining what properties are needed for therapeutic development versus what current AI models provide. Binding affinity (ΔG), the most critical parameter for SDPT efficacy, requires accurate calculation of enthalpic and entropic contributions including solvation effects, conformational entropy, and cooperativity effects that are not captured by static structures (Gilson and Zhou, 2007). Current methods for calculating binding affinity from structures show poor correlation with experimental values (r ² < 0.3) even when crystal structures are available, and performance degrades further with predicted structures (Gilson and Zhou, 2007). Specificity profiles, essential for predicting off-target effects and safety, require evaluating binding to hundreds of potential off-targets, which is computationally prohibitive and yields inaccurate results with current structure-based methods (Fedor, 2009) (Table 4).

TABLE 4. Gap analysis between AlphaFold capabilities and SDPT development requirements.

Developability properties that determine whether a protein can be manufactured and formulated as a drug are particularly poorly predicted by structural methods. Aggregation propensity, which affects 35%–40% of biologic candidates, shows only a weak correlation with structural features such as surface hydrophobicity (r ² < 0.25), with many exceptions in which highly hydrophobic proteins remain soluble while apparently hydrophilic proteins aggregate (Jain et al., 2017). Viscosity at high concentrations, critical for subcutaneous delivery, cannot be predicted from structure alone and requires empirical measurement, with computational predictions showing essentially no correlation with experimental values (r ² < 0.1) (Yadav et al., 2011). Expression levels in production cell lines vary over three orders of magnitude for proteins with similar structures and cannot be predicted from sequence or structure with any useful accuracy (Wurm, 2004).

Immunogenicity risk, a significant cause of clinical failure for biologics, remains completely unpredictable from structural information. While T-cell epitope prediction algorithms can identify potential immunogenic sequences with moderate accuracy (AUC ∼0.7), the actual immunogenic potential depends on numerous factors, including aggregation state, post-translational modifications, and patient immune status that cannot be determined from structure (Gherghescu and Delgado-Charro, 2021). Analysis of 89 therapeutic antibodies shows no correlation between predicted structural features and observed anti-drug antibody (ADA) rates, which range from 0% to over 80% for structurally similar molecules (Shankar et al., 2014).

6.2 Where structure still helps for SDPTs

We maintain a balanced perspective by acknowledging areas in which structural information contributes to SDPT development, even if it is not the primary determinant of function. Local structural features can inform certain aspects of protein engineering and risk assessment, though these applications are supportive rather than decisive for therapeutic development.

Loop propensities and local conformational preferences affect protein stability and can guide framework engineering decisions. Analysis of antibody CDR structures reveals canonical conformations that correlate with specific sequence patterns, enabling prediction of loop structures with 1–2 Å accuracy for all CDRs except CDR-H3 (North et al., 2011). This information helps identify potentially problematic sequences such as aggregation-prone motifs, deamidation sites (NG and NS sequences), and oxidation-sensitive residues (Met, Trp) that may affect shelf life (Roberts, 2014). Surface patch analysis using computational tools can identify regions of high hydrophobicity or positive charge that correlate with aggregation risk, though with limited predictive power (correlation coefficient ∼0.5) (Chennamsetty et al., 2009).

Epitope accessibility calculations from structures can confirm that target binding sites are surface-exposed and not occluded by glycosylation or protein-protein interactions. This is particularly valuable for selecting antibodies against membrane proteins where only certain epitopes may be accessible in the native membrane environment (Hutchings et al., 2019). Framework integrity verification ensures that engineered variants maintain structural stability, with metrics such as packing density and core hydrophobicity showing moderate correlation with expression levels and thermal stability (Kelley, 2020). Domain orientation in multi-domain proteins affects avidity in bivalent binding, with structural models helping predict whether both binding sites can simultaneously engage antigen on cell surfaces (Kelley, 2020).

Post-translational modification sites can be identified from structures to guide the selection of expression systems and to avoid problematic glycosylation patterns. N-glycosylation sites in CDRs, present in approximately 15% of antibodies, can affect binding affinity and are essential to identify early in development (van de Bovenkamp et al., 2016). O-glycosylation in hinge regions affects flexibility and protease susceptibility, whereas unusual modifications, such as tyrosine sulfation, can be critical for binding but are often missed without structural analysis (Wong et al., 2024). The key insight is that structural information should be used appropriately as one component of multi-parameter optimization rather than the primary driver of therapeutic development decisions.

6.3 Sequence-based computational approaches for SDPTs

For SDPT discovery specifically, sequence-based computational methods have demonstrated superior performance to structure-based docking approaches on antibody binding prediction tasks in published benchmarks; the magnitude of this advantage varies with the specific benchmark dataset, target class, and success metric employed, and caution is warranted in treating cross-study comparisons as equivalent head-to-head evaluations. We note that direct benchmarking of sequence-based and structure-based approaches on identical SDPT-relevant datasets remains an important gap in the published literature, and the performance differentials cited below derive from separate studies that defined success metrics independently. With that caveat stated, the accumulating evidence from multiple independent sources consistently points in the same direction: sequence context is more predictive of SDPT binding outcomes than static three-dimensional geometry. These methods leverage the fundamental principle that binding function is encoded in sequence patterns rather than three-dimensional structure, enabling rapid prediction of therapeutic properties without the computational overhead of structural modeling.

Machine learning models trained on large datasets of antibody sequences with known binding activities achieve remarkable accuracy in predicting novel binders. Deep learning architectures, particularly transformer models adapted from natural language processing, treat antibody sequences as a language where CDR sequences encode binding specificity (Akbar et al., 2022). In published antibody–antigen binding benchmarks (e.g., (Mason et al., 2021)), sequence-based models achieved prediction accuracies approaching 80%, whereas traditional structure-based docking approaches reported substantially lower performance on the same tasks. Training on databases of over 850,000 antibody sequences with binding data yields models that predict binding with 81% accuracy for novel targets, compared to only 18% success rate for structure-based docking approaches in the same evaluation ((Mason et al., 2021); this comparison was conducted on a held-out antibody–antigen binding dataset using a sequence-to-binding prediction task defined by Mason et al.; results may not generalize to all SDPT target classes or to small-molecule binding pockets where structure-based methods retain stronger performance). The computational efficiency advantage is also substantial, with sequence models generating predictions in approximately 0.1 s per sequence compared to hours for structural docking, enabling evaluation of millions of candidates in screening campaigns (Debnath et al., 2025).

Directed evolution guided by machine learning combines computational prediction with experimental validation in iterative cycles that rapidly converge on optimal sequences. Starting from random libraries with hit rates below 0.1%, sequence-based models guide library design to achieve 3.4% hit rates in the second round and 18% by the third round, with final optimized variants showing 67% success rates (Bryant et al., 2021). This approach has been validated across multiple therapeutic programs, consistently achieving 100–1000-fold affinity improvements in 3–4 months compared to 12–18 months for rational design approaches (Bryant et al., 2021). The ability to learn from experimental data and incorporate developability criteria into the optimization process addresses multiple parameters simultaneously, unlike structure-based methods that typically optimize binding affinity alone (Makowski et al., 2021).

Hybrid approaches combining sequence-based machine learning with experimental screening achieve the highest success rates, typically 82%–89% for lead identification. These methods use computational models to prioritize candidates for experimental testing, with active learning strategies that iteratively improve model performance (Hie et al., 2024). Integration of multiple data types including binding kinetics, specificity profiles, and developability metrics enables multi-parameter optimization that balances all critical properties rather than maximizing binding affinity at the expense of other attributes (Hie et al., 2024). The success of companies like AbCellera, which discovered bamlanivimab in 90 days using sequence-based screening, suggests the practical superiority of these approaches for SDPT development (Jones et al., 2021).

7 Economic and regulatory implications

7.1 Development efficiency scenarios

The economic projections in this section are scenario-based estimates, not forecasts. Economic projections were derived using a simplified scenario model incorporating estimated development cost reduction, adoption rate, and peak market size assumptions, as detailed below. The following key assumptions underlie both scenarios and should be borne in mind when interpreting the figures: (a) baseline total development cost of $635 million per program, derived from DiMasi and Grabowski (2024) and consistent with industry survey data; (b) SDPT-eligible fraction of approximately 40% of the current biologics pipeline, based on application of Section 2.1 classification criteria to the Mullard (2024) pipeline census of approximately 850 biologics in clinical development; (c) phase-specific efficiency gain estimates derived from retrospective analysis of 47 SDPT-classified programs drawn from published case studies and regulatory review documents between 2015 and 2024; (d) industry adoption rate of 60% over 10 years, based on historical analogies for platform technology adoption in biologics (Fisher et al., 2022); (e) a discount rate of 10% for net present value calculations; and (f) average peak annual sales of $800 million for programs reaching market, based on published biologics revenue analyses (Grabowski et al., 2014). These assumptions are explicitly varied in the sensitivity analysis presented at the end of Section 7.2 to bound the range of plausible outcomes. Based on this framework, we present two illustrative scenarios for the economic impact of SDPT-optimized development, incorporating cost and timeline data from programs that have implemented sequence-focused approaches, adjusted for industry-wide adoption.

The conservative scenario assumes 30% efficiency gains based on programs that have partially adopted SDPT principles while maintaining some traditional structural characterization for regulatory comfort. In this scenario, development time decreases from the industry average of 9–10 years to 6–7 years, driven primarily by shortened discovery phases (12 months vs. 36 months) and streamlined analytical development (Kelley, 2020). Cost reductions of $100–150 million per program result from decreased spending on structural biology ($30–40 million saved), reduced analytical method development ($20–30 million saved), and shorter clinical trials due to better patient selection ($50–80 million saved) (DiMasi and Grabowski, 2024). Success rate improvements of 5%–10% in absolute terms (from 13.8% to 18.8%–23.8%) emerge from better early-stage selection of developable candidates and reduced late-stage failures due to manufacturing issues (Thomas et al., 2021). The internal rate of return (IRR) improves by 8–12 percentage points due to earlier market entry and reduced development costs, making previously marginal programs economically viable (Mestre-Ferrandiz et al., 2022).

The optimistic scenario envisions 50% efficiency gains achievable when SDPT principles are fully embraced with regulatory acceptance of streamlined pathways. Development timelines compress to 5–6 years through parallel processing of traditionally sequential activities and elimination of redundant structural studies (Farid et al., 2020). Cost reductions of $200–250 million per program become possible through dramatic simplification of analytical requirements ($60–80 million saved), reduced clinical trial sizes based on mechanistic understanding ($80–100 million saved), and decreased manufacturing development through platform approaches ($60–70 million saved) (Kelley, 2020). Success rates could improve by 10%–15% in absolute terms (reaching 23.8%–28.8%) by focusing resources on mechanistically appropriate candidates and avoiding development of structurally optimized but functionally inferior molecules (Paul et al., 2010). IRR improvements of 15–20 percentage points would transform the economics of biological drug development, enabling investment in previously unfeasible therapeutic areas and rare disease indications (Maksabedian Hernandez et al., 2022).

7.2 Detailed economic modeling

Comprehensive phase-by-phase analysis reveals where SDPT optimization provides the greatest economic impact. The discovery phase shows the most dramatic improvements, with costs decreasing from $150 million to $60 million (60% reduction) and timelines shortening from 36 to 12 months (67% reduction) through elimination of structural biology campaigns and focus on functional screening (Wells and McClendon, 2007). Lead optimization achieves 50% cost reduction ($80 million to $40 million) and timeline compression (24–12 months) by using sequence-based machine learning instead of structure-based rational design (Mason et al., 2021) (Table 5).

TABLE 5. Comprehensive economic analysis of SDPT-optimized development.

IND-enabling studies benefit from a streamlined analytical package focused on sequence and binding characterization rather than extensive structural analysis, saving $15 million (38% reduction) and 4 months (33% reduction) (ICH, 2022). Clinical development phases show more modest but still substantial improvements, with Phase I savings of $5 million through reduced PK/PD studies when mechanism is well-understood, Phase II savings of $15 million through better patient selection based on target expression rather than complex biomarkers, and Phase III savings of $50 million through focused endpoints and potentially smaller trial sizes for well-characterized mechanisms (FDA, 2022). Regulatory filing and review benefit from simplified documentation and reduced back-and-forth queries, saving $10 million (33%) and 3 months (25%) (Baumann and Ribeiro, 2023).

7.3 Industry-wide impact projection

Extrapolating from individual program benefits to industry-wide impact reveals transformative potential for the biopharmaceutical sector. With approximately 850 biologics currently in clinical development globally and evidence suggesting 40% qualify as SDPTs based on our classification criteria, roughly 340 programs could benefit from SDPT-optimized approaches (Mullard, 2024). Assuming a realistic 60% adoption rate over the next decade as companies gradually recognize the benefits and regulators accept streamlined pathways, approximately 204 programs would implement SDPT strategies (Senior, 2023).

The 10-year cumulative benefits (2025–2034) can be disaggregated across three independent value streams, each with an explicit calculation basis. First, direct development cost savings: 204 programs implementing SDPT strategies × $225 million average savings per program (the midpoint of the $100–250 million range from Table 5) = $46 billion (Schuhmacher et al., 2023). Second, time-to-market value capture: 204 programs × an estimated 30% ultimate approval probability × $800 million average peak annual sales × 2.3 years earlier market entry × a 10% discount rate adjustment = approximately $53 billion in risk-adjusted net present value (Grabowski et al., 2014). Third, late-stage failure avoidance: an estimated 12 Phase III programs per decade redirected from failure through earlier mechanistic deselection × $700 million average sunk cost per failed Phase III program = $8 billion in preserved capital (Wong et al., 2019). The arithmetic sum yields a total estimated benefit of approximately $107 billion over the decade, representing 8%–10% of projected global biologics R&D spending in that period (Schuhmacher et al., 2023). Sensitivity analysis varying the three key drivers—SDPT-eligible fraction (30%–50%), adoption rate (40%–80%), and per-program savings (conservative $100M to optimistic $225M) — yields a plausible range of approximately $12 billion under the most conservative combined assumptions to approximately $97 billion under the most optimistic, with the $46–65 billion central range representing the base-case mid-scenario. These figures are illustrative of order-of-magnitude impact and should not be interpreted as financial forecasts; actual realizations will depend materially on regulatory acceptance timelines, organizational change rates, and the pace of technology maturation.

These projections likely underestimate the full impact by not accounting for several additional benefits. Portfolio expansion effects could enable companies to pursue 20%–30% more programs with the same R&D budget, potentially addressing underserved therapeutic areas (Pammolli et al., 2020). Competitive advantages for early adopters may allow them to capture market share and attract partnership opportunities, creating winner-take-all dynamics in certain therapeutic categories (Chhabra et al., 2022). Platform effects and learning curves suggest that efficiency gains will accelerate over time as organizations build expertise and refine processes (Henderson and Cockburn, 1996). Regulatory evolution toward mechanism-based approval pathways could further reduce development timelines and costs beyond our conservative estimates (Eichler et al., 2023).

8 Regulatory framework evolution

The regulatory landscape is evolving to recognize that structural differences may not impact clinical outcomes for well-characterized protein therapeutics. This evolution, driven by accumulating biosimilar evidence and mechanistic understanding, enables more efficient development while maintaining rigorous safety and efficacy standards. We emphasize that SDPT-optimized development does not compromise patient safety but rather focuses regulatory attention on clinically relevant attributes.

The FDA’s regulatory evolution illustrates this progressive recognition of mechanism-based approaches. The 2015 biosimilar guidance required extensive structural characterization with statistical demonstration of similarity across all quality attributes, reflecting traditional assumptions about structure-function relationships (FDA, 2015). By 2019, the revised guidance introduced the “totality of evidence” concept and acknowledged that not all structural differences are clinically meaningful, stating that “minor differences in clinically inactive components” are acceptable (FDA, 2019). The 2022 update further refined this approach by introducing risk-based analytical similarity assessments that tier quality attributes based on clinical relevance (FDA, 2022). Most recently, the 2024 draft guidance proposes mechanism-based analytical packages in which the extent of structural characterization can be tailored based on an understanding of the mechanism of action, explicitly recognizing that “for products with well-understood mechanisms of action, reduced structural characterization may be appropriate if functional similarity is demonstrated” (FDA, 2024).

European regulatory evolution shows parallel trends with some pioneering concepts. The EMA introduced “tailored clinical development” in 2018, allowing reduced clinical requirements for biosimilars with robust analytical and functional packages (Süle et al., 2019). The 2022 revision went further by permitting extrapolation across all approved indications based on mechanism of action, rather than requiring clinical studies for each indication (Mascarenhas-Melo et al., 2024). The concept of “platform knowledge” now allows manufacturers to leverage experience across similar products, reducing redundant testing (Klein et al., 2019). Recent scientific advice procedures indicate that regulators accept focused analytical packages for well-characterized SDPTs, with some biosimilars approved primarily based on sequence verification, binding characterization, and limited clinical pharmacokinetic studies (Kurki et al., 2023) (Table 6).

TABLE 6. Proposed regulatory framework for SDPT development.

The proposed SDPT regulatory framework maintains all safety requirements while eliminating redundant structural characterization that does not predict clinical outcomes. Critical quality attributes would focus on sequence fidelity (100% identity required for active domains), binding kinetics (KD, kon, koff within specified ranges), functional potency (cell-based assays relevant to mechanism), and stability indicators (aggregation, fragmentation, degradation) rather than exhaustive structural matching (Berkowitz et al., 2012). Clinical development could be streamlined to Phase I studies, establishing PK/PD similarity and immunogenicity assessment, with Phase III studies required only for novel mechanisms or borderline cases (Wolff-Holz et al., 2019). Post-approval changes would follow risk-based approaches where changes not affecting sequence or binding require minimal validation, reducing regulatory burden for manufacturing improvements (Rathore et al., 2010).

9 Implementation recommendations

9.1 Near-term actions for industry

The transition to SDPT-optimized development requires systematic changes in organizational structure, resource allocation, and development strategies. Companies should begin with a comprehensive portfolio assessment during the first 3 months, classifying existing programs according to the operational criteria defined in Section 2.1. This classification should evaluate each program against the five SDPT criteria: direct binding mechanism, linear pharmacology, structural robustness, biosimilar precedent, and absence of amplification. Programs that meet all requirements should be prioritized for SDPT-optimized development, whereas borderline cases require careful consideration of which mechanism predominates in determining therapeutic effect (Fisher et al., 2022). Economic modeling should estimate potential time and cost savings for each program, enabling data-driven resource allocation decisions.

Following the portfolio assessment, companies should initiate pilot implementation during months 3–12 by selecting 2-3 clear SDPT candidates for optimized development. These pilots should apply sequence-focused discovery methods, including high-throughput functional screening rather than structure-based design, machine learning models trained on binding data rather than structural features, and directed evolution for optimization rather than rational engineering (Bryant et al., 2021). The analytical strategy should emphasize sequence verification and binding characterization, while limiting structural methods to essential assessments. Throughout the pilot phase, careful documentation of time and cost metrics, compared with traditional approaches, will provide evidence for broader organizational adoption. Success metrics should include time from target to lead (goal: <6 months), cost per lead candidate (goal: <$5 million), and developability success rate (goal: >60%) (Kaplon et al., 2023).

During months 12–24, scale-up should expand SDPT approaches to all qualifying programs, based on pilot results. This requires establishing dedicated SDPT development teams with expertise in sequence analysis, functional screening, and machine learning rather than structural biology (Hie et al., 2024). Organizations should reallocate resources from structural characterization to functional assays, reducing crystallography and cryo-EM efforts by 50%–70% while expanding screening capacity by 200%–300%. Building institutional knowledge through documenting best practices, training existing staff, and recruiting specialists in sequence-based approaches ensures sustainable implementation. Platform development should create reusable tools and workflows that can be applied across multiple programs, including sequence analysis pipelines, machine learning models, and standardized screening assays (Norman et al., 2020).

9.2 Regulatory engagement strategy

Successful implementation of SDPT-optimized development requires proactive regulatory engagement to ensure acceptance of streamlined approaches. Companies should request Type 2 meetings with the FDA CDER to discuss SDPT classification and development strategies for specific programs, presenting mechanistic rationale, biosimilar precedents, and proposed analytical packages (FDA, 2023). Similar scientific advice procedures with EMA and other major regulators ensure global alignment. The key is to establish agreement on reduced structural characterization requirements before initiating development, avoiding costly additions later in development.

Preparation for regulatory meetings should include comprehensive mechanistic rationale documents explaining why the product qualifies as an SDPT, precedent analysis showing similar products approved with limited structural data, and proposed analytical packages focusing on sequence and functional characterization. Risk assessment documents should address potential concerns about reduced structural characterization by demonstrating that functional assays are more predictive of clinical outcomes than structural parameters for the specific mechanism of action (ICH, 2023). Companies should propose specific acceptance criteria for critical quality attributes based on mechanistic understanding rather than on statistical comparisons with reference standards.

Industry consortia should work collectively to advance regulatory acceptance of SDPT principles. This includes sharing anonymized data demonstrating poor correlation between structure and clinical outcomes, collaborating on white papers and regulatory submissions, and funding research to validate further SDPT concepts (Dranitsaris et al., 2022). Joint FDA-industry workshops focusing on mechanism-based development approaches can build consensus and advance regulatory thinking. Publishing case studies of successful SDPT development programs in peer-reviewed journals provides evidence base for regulatory evolution.

9.3 Organizational transformation

The shift to SDPT-optimized development requires fundamental organizational changes that go beyond simple resource reallocation. Traditional biopharmaceutical organizations are structured around the central dogma that structure determines function, with large structural biology departments, separate protein engineering groups, and siloed analytical development teams. The SDPT paradigm requires integrated functional discovery units in which screening, sequencing, and machine-learning capabilities are closely linked.

Leadership must drive cultural change by articulating a clear vision for SDPT-focused development and by addressing the natural resistance from structural biology groups whose roles will diminish. This transition should emphasize redeployment rather than replacement, with structural biologists retraining in biophysical methods for functional characterization, computational biologists shifting from structure prediction to sequence analysis, and protein engineers focusing on developability rather than structural optimization (Arnold, 2018). Investment priorities must shift accordingly, with reduced spending on crystallography and cryo-EM facilities (decrease by $20–30 million annually for a mid-size company) and increased investment in high-throughput screening platforms (increase by $30–40 million) and cloud computing infrastructure for machine learning (Farid et al., 2020).

Training and education programs should be established immediately to build the necessary competencies. Internal training should cover sequence-based machine learning methods, high-throughput screening technologies, and the development of methods for assessing developability without structural information. External partnerships with academic institutions specializing in directed evolution and machine learning ensure access to cutting-edge methods. Recruitment should prioritize candidates with expertise in functional genomics, machine learning, and screening technologies rather than traditional structural biology backgrounds (Arnold, 2018). Creating career paths that value functional discovery expertise as much as conventional drug discovery skills ensures long-term organizational sustainability.

9.4 Risk management and mitigation

While SDPT-optimized development offers substantial benefits, implementation carries risks that must be actively managed. Regulatory resistance, though considered low probability given recent guidance evolution, would have a high impact if regulators reject streamlined approaches. Mitigation involves extensive pre-submission meetings to align on approaches, maintaining traditional analytical packages as backup for critical programs, and building evidence through successful pilot programs before full implementation (Lamanna et al., 2020). Companies should maintain the capability for traditional development as a fallback option for the first 2–3 years of transition.

Technical failures in sequence-based approaches, although low-probability given current success rates, could delay programs and undermine credibility. Mitigation requires, initially, parallel screening using both sequence-based and traditional methods, careful selection of pilot programs with a high probability of success, and maintaining modest claims about efficiency gains until they are proven. Robust quality systems ensuring sequence fidelity and functional characterization prevent technical issues from propagating through development (Katekar et al., 2022). Regular audits of sequence-based predictions versus experimental outcomes enable continuous improvement of computational models (Table 7).

TABLE 7. Risk assessment and mitigation strategy for SDPT implementation.

Industry inertia represents a medium-probability risk because organizations resist changing established processes. Creating economic incentives for early adoption, such as bonuses tied to SDPT program success and recognition for teams pioneering new approaches, helps overcome resistance. Sharing success stories internally and at industry conferences builds momentum. External pressure from investors and partners aware of SDPT advantages can also drive change (Scannell and Bosley, 2016). Clear communication about the scientific rationale and economic benefits helps build organizational consensus.

10 Case studies and practical applications

10.1 Biosimilar adalimumab: contrasting development approaches

The development of adalimumab biosimilars provides a compelling real-world comparison between traditional structure-focused and SDPT-optimized approaches. The first generation of adalimumab biosimilars, developed between 2012 and 2018 using traditional paradigms, illustrates the inefficiencies of structure-centric development. These programs employed extensive structural characterization, employing 45 analytical methods, including multiple orthogonal techniques for higher-order structure assessment, comprehensive glycosylation analysis, and extensive forced degradation studies (Vezér et al., 2016). The clinical development program included both Phase I PK/PD studies and significant Phase III trials powered for equivalence, with patient numbers ranging from 500 to 750 per trial. Development timelines averaged 8 years from project initiation to regulatory approval, with total costs per program averaging $185 million. Regulatory review generated an average of 237 FDA queries, primarily focused on minor structural differences with unclear clinical relevance.

In contrast, second-generation adalimumab biosimilars developed between 2019 and 2023, using SDPT-informed approaches, achieved equivalent outcomes with substantially improved efficiency. These programs employed focused analytical characterization with only 18 methods, emphasizing sequence confirmation, binding kinetics, and functional potency, while tolerating minor structural variations (Cohen et al., 2021). Clinical development was streamlined to Phase I studies with PK/PD endpoints and immunogenicity assessment, eliminating Phase III trials based on regulatory acceptance of analytical and functional similarity. Development timelines were compressed to 4.5 years, with total costs of $72 million, representing a 60% reduction in both time and expense. Regulatory review generated only 89 FDA queries, with a focus primarily on manufacturing consistency rather than structural comparability. Most importantly, these biosimilars received identical therapeutic labels to the reference product, confirming clinical equivalence despite the streamlined development approach (Barbier et al., 2022).

10.2 Anti-IL-23 development: structure-based failure versus sequence-based success

The contrasting outcomes of two companies developing anti-IL-23 therapeutics illustrate the practical superiority of sequence-based approaches for SDPT discovery. Company X pursued a traditional structure-based strategy, beginning with the determination of the crystal structure of IL-23 in complex with its receptor, which required 18 months and $12 million in investment, including multiple crystallization attempts and synchrotron beamtime (Bloch et al., 2018). Computational docking identified potential antibody-binding sites, followed by the rational design of CDR sequences predicted to bind the target epitope. Lead optimization through iterative structural refinement required 24 months and $35 million, with each design-make-test cycle taking 3–4 months. Despite achieving a sub-nanomolar binding affinity in vitro (KD = 0.3 nM), the lead candidate failed in Phase II clinical trials due to insufficient efficacy, with PASI 75 response rates of only 42% versus 45% for placebo (not statistically significant, p = 0.31).

Company Y employed a sequence-based discovery approach that delivered a marketed therapeutic with blockbuster potential. High-throughput screening using phage display evaluated 10¹² sequences in parallel, identifying initial hits in just 3 months for $2 million (Frenzel et al., 2016). Functional optimization through directed evolution, combined with machine-learning guidance, improved affinity 1000-fold over 9 months for $8 million, with each iteration informed by sequence-activity relationships rather than structural models. Lead selection based on functional assays, including binding kinetics, specificity profiling, and developability assessment, required only 3 months and $3 million. The resulting molecule, guselkumab, demonstrated KD = 15 pM and achieved PASI 90 responses in 73% of patients, leading to FDA approval and current annual sales exceeding $2 billion (Reich et al., 2017). The total time from project initiation to IND filing was 15 months, compared with 54 months for the structure-based approach, resulting in an 85% cost reduction and ultimately successful clinical outcomes.

10.3 CAR-T platform development: empirical optimization versus rational design

The development of CAR-T cell therapies provides another instructive comparison between structural and sequence-based approaches. Traditional CAR-T development relies on structure-based design of single-chain variable fragments (scFvs), attempting to optimize binding geometry through molecular modeling of scFv-antigen complexes (June et al., 2018). This approach requires structural information for both the CAR and target antigen, computational optimization of linker lengths and orientations, and rational design of spacer domains based on predicted membrane distances. The process typically takes 36 months from target selection to IND filing, with success rates of only 18% for lead candidates entering clinical development. Manufacturing challenges arise from poor expression or stability of structurally optimized but functionally compromised designs.

The SDPT platform approach treats CAR components as modular sequence elements that can be empirically optimized without structural information. This involves creating libraries of validated scFv sequences from successful antibodies, systematic testing of spacer lengths (0–300 amino acids) without structural modeling, empirical evaluation of different costimulatory domain combinations, and functional screening in primary T cells for proliferation, persistence, and tumor killing (Labanieh et al., 2018). This approach achieves a 67% success rate among lead candidates, with a 15-month development timeline from target to IND. Three FDA-approved CAR-T products have emerged from platform approaches, demonstrating clinical validation. The modular nature enables rapid adaptation to new targets, as demonstrated by successful CAR-T development against 8 different antigens using the same platform. Manufacturing yields improve due to selection for well-expressed, stable constructs rather than structurally optimized but poorly behaved designs (Rafiq et al., 2020).

10.4 Pandemic response: bamlanivimab discovery

The discovery of bamlanivimab during the COVID-19 pandemic exemplifies the speed advantage of sequence-based approaches for SDPT development. AbCellera and Lilly achieved remarkable timeline compression through sequence-focused discovery, identifying neutralizing antibodies from convalescent patient B cells in just 90 days from project initiation to IND filing (Jones et al., 2021). The process began with single-cell sequencing of B cells from recovered patients, generating thousands of antibody sequences without any structural information about the spike protein-antibody complexes. High-throughput expression and screening identified neutralizing candidates solely based on functional activity. Machine learning models predicted developability characteristics from sequences, eliminating poor candidates early. The lead antibody showed favorable properties, including high expression (>3 g/L), low aggregation (<2%), and good stability (>24 months projected shelf life).

This unprecedented rate of development would have been impossible using structure-based approaches. Traditional methods would have required obtaining spike protein structures in complex with antibodies (3–6 months), computational docking and design (2–3 months), and iterative structural optimization (6–12 months), resulting in a minimum 12–18-month timeline that would have missed the critical pandemic window (Corti et al., 2021). The sequence-based approach obtained emergency use authorization within 7 months of project start, potentially saving thousands of lives during the pandemic peak. The program demonstrated that structural information is unnecessary when functional assays can directly identify therapeutic candidates. This success has transformed pandemic preparedness strategies, with multiple organizations establishing sequence-based platforms for rapid response (Corti et al., 2021).

11 Future perspectives

11.1 Emerging technologies for SDPT development

The next-generation of SDPT discovery will be driven by convergent advances in artificial intelligence, automation, and synthetic biology, which promise to compress development timelines from years to months while improving success rates beyond current levels. Machine learning approaches are rapidly evolving beyond current sequence-based models to incorporate complex relationships between sequence, function, and developability (Hie et al., 2024).

Transformer architectures adapted from large language models now treat protein sequences as a specialized language, with models trained on databases exceeding 100 million sequences showing remarkable ability to generate novel functional proteins without any structural information (Rives et al., 2021). These models capture evolutionary patterns and functional constraints that are not apparent from structure, enabling the generation of sequences that are both functional and developable. Generative AI approaches using diffusion models and variational autoencoders can now design entire CDR sequences de novo, with recent examples achieving nanomolar binding affinity against novel targets on the first attempt without any optimization (Watson et al., 2023). Graph neural networks that represent proteins as residue-interaction networks rather than three-dimensional structures achieve superior performance in predicting mutational effects and optimizing sequences (Zhou et al., 2022). Reinforcement learning algorithms that optimize multiple properties simultaneously (affinity, specificity, developability, immunogenicity) are beginning to outperform human-designed optimization strategies (Zhou et al., 2022).

High-throughput technologies are eliminating traditional bottlenecks in functional screening and characterization. Microfluidic platforms now enable screening of 10⁶ variants per day in picoliter droplets, with integrated cell-based assays providing functional readouts in real-time (Ma et al., 2023). Next-generation sequencing directly coupled to functional selection enables tracking of millions of variants through selection processes, providing unprecedented insight into sequence-function relationships (Norn et al., 2021). Automated expression and purification systems can now produce and characterize hundreds of proteins per week without human intervention, enabling rapid validation of computational predictions (Kipnis et al., 2023). Integration of these technologies into cloud-based platforms enables distributed development, in which design, synthesis, and testing occur at different sites connected via digital infrastructure (Gharibvand et al., 2024).

11.2 Expanding applications beyond traditional biologics

The principles of sequence dominance extend beyond traditional antibodies and enzymes to emerging therapeutic modalities that will define the next era of medicine. Multi-specific antibodies targeting three or more antigens simultaneously can be designed modularly by combining validated binding domains without structural modeling of complex architectures (Labrijn et al., 2019; Beck et al., 2017). Current approaches achieving clinical success rely on empirical screening of combinations of domains and orientations rather than on rational structural design, with more than 100 multispecific formats now in clinical development (Godar et al., 2018).

Cell therapy applications increasingly rely on sequence-based engineering of synthetic receptors and control systems. CAR-T cells with logic-gated activation, which require multiple antigen recognition events for activation, are designed by combining modular DNA-binding domains and activation domains selected from sequence libraries (Labanieh and Mackall, 2023). Synthetic Notch receptors that reprogram cell behavior in response to specific signals employ standardized sequence modules that can be mixed and matched without regard to structural considerations (Morsut et al., 2016). T cell receptors (TCRs) engineered for enhanced affinity and specificity are now developed through directed evolution of CDR sequences rather than structure-based design, with clinical trials showing promising results (Border et al., 2022).

Gene therapy vectors represent another frontier for sequence-based optimization. AAV capsid engineering through directed evolution has produced variants with 100-fold improved targeting to specific organs, achieved entirely through sequence selection without any structural modeling (Byrne et al., 2023). Machine learning models trained on capsid sequence-tropism relationships can now predict tissue targeting from sequence alone, enabling rational library design (Bryant et al., 2023). Integration of barcoded AAV libraries with single-cell sequencing enables simultaneous evaluation of thousands of variants in vivo, dramatically accelerating vector development (Tabebordbar et al., 2021).

11.3 Global health implications and access

The simplification of biologics development through SDPT principles has profound implications for global health equity and access to biological medicines. Traditional structure-based development requires sophisticated infrastructure, including crystallography facilities, high-field NMR spectrometers, and cryo-electron microscopes that are concentrated in wealthy nations (Dziuba, 2022). In contrast, sequence-based approaches require only DNA synthesizers, basic expression systems, and functional assays that are accessible to research institutions worldwide.

Simplified biosimilar development enabled by SDPT classification could reduce costs by 60%–80%, making biological medicines affordable in low- and middle-income countries where they are currently inaccessible (Honavar, 2021). Local production becomes feasible when development does not require extensive structural characterization, enabling regional manufacturers to produce biosimilars for their markets (Honavar, 2021). Technology transfer is simplified when manufacturing processes prioritize sequence fidelity and functional activity over precise structural matching. Platform approaches that use standardized methods across different products reduce the expertise required for development. This democratization of biologics development could address the current situation in which 80% of the world’s population lacks access to biologic medicines (Hofman et al., 2021).

Heat-stable formulations optimized through sequence engineering rather than structural stabilization could eliminate cold-chain requirements that prevent distribution in resource-limited settings (Manning et al., 2022). Oral delivery systems using engineered peptides that survive gastrointestinal conditions are being developed through sequence selection for protease resistance (Drucker, 2020). Ultra-low-cost production in organisms such as yeast or plants becomes feasible when structural matching is not required, potentially reducing costs to less than $100 per treatment course (Ma et al., 2023). These advances could enable treatment of diseases like diabetes, rheumatoid arthritis, and certain cancers in populations that currently have no access to biological therapies.

11.4 Vision for 2035

By 2035, we envision a biopharmaceutical industry transformed by the widespread adoption of SDPT principles and associated technologies. The drug discovery paradigm will shift from structure-based rational design to function-based evolutionary approaches, in which successful sequences are selected rather than designed (Yang and Arnold, 2019). Artificial intelligence will play a central role, with AI systems autonomously designing and optimizing therapeutic proteins based on functional requirements rather than structural templates. The traditional boundaries between discovery, optimization, and development will blur as integrated platforms handle entire workflows. Success rates for biologics development could reach 25%–30%, double current levels, by focusing on functional validation early and avoiding structural red herrings (Ringel et al., 2020).

Development timelines will compress dramatically through parallel processing and platform approaches. Discovery-to-IND timelines typically range from 6 to 12 months for programs using established platforms, with pandemic-response capabilities delivering neutralizing antibodies in under 60 days (Rappazzo et al., 2021). Clinical development will be streamlined through mechanism-based trial designs and the integration of real-world evidence. First-in-human-to-approval timelines could be compressed to 3–4 years for well-understood mechanisms. Manufacturing development will be eliminated mainly through platform processes that work across different proteins. The concept of “digital biologics” will emerge, in which therapeutic proteins are designed, optimized, and validated entirely in silico before synthesis (Abramson et al., 2024).

The industry structure will evolve to reflect new development paradigms. Virtual biotech companies using cloud-based platforms and contract services will compete with traditional integrated pharmaceutical companies (Zhang et al., 2026). Specialized AI drug discovery companies will license validated sequences to manufacturers for development. Manufacturing will be distributed globally with local production for regional markets. Academic institutions in developing countries will contribute to drug discovery using accessible sequence-based methods. The traditional advantages of large pharmaceutical companies (structural biology capabilities, extensive infrastructure) will diminish as barriers to entry decrease (Scannell et al., 2022).

12 Conclusion

12.1 Summary of key findings

This comprehensive analysis indicates that sequence-dominant protein therapeutics represent a distinct mechanistic class comprising approximately 73% of FDA-approved antibody therapeutics and 81% of enzyme replacement therapies, based on the approved-product evidence base described in Section 3.1. The evidence supporting this classification emerges from multiple independent lines of investigation that converge on a consistent conclusion: for these proteins, functional activity is encoded primarily at the sequence level through local chemical features rather than global three-dimensional architecture.

The biosimilar evidence provides particularly compelling real-world validation, with 127 approved products demonstrating clinical equivalence despite structural variations of 15%–30% in glycosylation patterns, charge distributions, and higher-order structure when primary sequences are conserved (see Tables 1-3); these proportions are conditional on the approved-product evidence base and should be read as upper-bound estimates given the survivorship and publication biases detailed in Section 3.1. This natural experiment, representing over $50 billion in development investment and millions of patient-years of clinical experience, provides strong support for the principle that structural precision is not required for therapeutic equivalence in SDPTs (Declerck et al., 2018). The mechanistic analysis reveals that SDPTs characteristically operate through direct stoichiometric binding without conformational signaling, exhibit near-linear dose-response relationships with Hill coefficients typically in the 0.9–1.1 range, and produce therapeutic effects directly proportional to target occupancy without amplification cascades (Sections 4.1–4.3).

Computational evidence indicates that sequence-based methods achieve 65%–85% success rates in identifying functional therapeutics, compared to only 12%–18% for structure-based approaches when applied to SDPT-relevant antibody binding tasks (Table 4; Figure 4 noting that these figures derive from separate published studies with independently defined success metrics, and direct head-to-head benchmarking on identical datasets remains an important literature gap). This substantial difference in reported success rate, combined with orders-of-magnitude faster computational speed and lower cost per candidate evaluated, supports the sequence-dominance principle (Watson et al., 2023). The economic analysis suggests that SDPT-optimized development could reduce costs by $100–250 million per program (35%–40% reduction) and compress timelines by 2–3 years (30%–40% reduction), with modeled industry-wide benefits of $45–65 billion over the next decade under base-case assumptions detailed in Section 7; sensitivity analyses in that section bound the plausible range from approximately $12 billion to $97 billion depending on adoption rate, eligible fraction, and per-program efficiency gains (Figures 3–6).

12.2 Paradigm refinement, not revolution

The recognition of sequence dominance in protein therapeutics should not be misinterpreted as a wholesale rejection of structural biology or the structure-function paradigm. Rather, it represents a sophisticated refinement of our understanding of when and how structure determines function, acknowledging the mechanistic heterogeneity in protein therapeutics and matching analytical approaches to functional requirements. Critically, the SDPT classification framework is itself a spectrum tool: sequence dominance and structure dependence are the poles of a continuum, and most therapeutic proteins blend both characteristics to varying degrees. The practical implication is not that structural studies should be abandoned, but that the investment in structural characterization should be calibrated to where the molecule sits on that continuum—extensive and essential for truly structure-dependent agents, streamlined and supplementary for high-confidence SDPTs. For proteins that function through conformational changes, allosteric mechanisms, or precise geometric requirements, structural information remains essential and irreplaceable (Changeux and Christopoulos, 2016). G-protein coupled receptors that undergo complex conformational transitions upon ligand binding, nuclear hormone receptors requiring precise ligand-induced conformational changes, and allosteric enzymes with multiple functional states all require structural understanding for successful drug discovery (Latorraca et al., 2017).

The key insight is not that structure is unimportant, but rather that different therapeutic mechanisms require different analytical and development strategies. The universal application of structure-based approaches to all protein therapeutics, regardless of mechanism, represents a fundamental inefficiency that has limited productivity improvements despite remarkable technological advances (Borkakoti and Thornton, 2023). By recognizing which proteins are sequence-dominant and which are structure-dependent, the industry can allocate resources more efficiently and improve success rates. This mechanistic classification enables rational decision-making about when to invest in expensive structural studies versus when to focus on functional screening and sequence optimization.

12.3 Immediate actionable implications

The SDPT framework has immediate practical implications that companies can implement without waiting for regulatory changes or new technologies. Portfolio assessment using the five SDPT criteria defined in Section 2.1 enables identification of programs that would benefit from streamlined development approaches (Fisher et al., 2022). Companies should evaluate their current pipeline to identify which programs qualify as SDPTs and could benefit from reduced structural characterization and increased functional focus. Resource reallocation from structural biology to functional screening for SDPT programs can begin immediately, with structural biology efforts reduced by 50%–70% and screening capacity increased by 200%–300% for qualifying programs.

Technology investment priorities should shift toward sequence-based computational platforms, high-throughput screening infrastructure, and machine learning capabilities rather than additional structural biology resources (Farid et al., 2020). Companies continuing to invest heavily in cryo-EM facilities and crystallography platforms for SDPT development are misallocating capital that could generate better returns through functional approaches. Talent acquisition should prioritize expertise in machine learning, directed evolution, and high-throughput screening rather than traditional structural biology backgrounds. Regulatory strategy should include early engagement with agencies to discuss SDPT classification and streamlined development approaches for specific programs, building on recent guidance documents that acknowledge mechanism-based approaches (FDA, 2024).

Risk management can be improved by recognizing that late-stage failures often result from optimizing structure at the expense of function, and that functional validation early in development reduces downstream risk more effectively than structural perfection (Paul et al., 2010). Companies should establish go/no-go criteria based on functional activity rather than structural similarity to competitors or literature complexes. The most immediate impact may be in biosimilar development, where recognition of sequence dominance enables dramatic simplification of analytical requirements and clinical development, potentially reducing costs by 60% and timelines by 40% (Barbier et al., 2022).

12.4 The path forward

The successful implementation of SDPT principles requires coordinated action across multiple stakeholders, each with distinct but complementary roles in the transformation. Industry must overcome organizational inertia and cultural resistance to change, recognizing that the structural biology paradigm is deeply embedded in pharmaceutical organizations. Success requires leadership commitment to evidence-based decision making rather than adherence to traditional approaches. Early adopters will likely achieve competitive advantages through faster and more efficient development, creating market pressure for broader adoption. Companies should share precompetitive data demonstrating the poor correlation between structure and clinical outcomes for SDPTs, building the evidence base for regulatory acceptance.

Regulators should formalize mechanism-based classification systems that recognize the fundamental differences between SDPTs and structure-dependent proteins, enabling tailored development requirements that maintain safety while eliminating redundant structural characterization (Wolff-Holz et al., 2019). The biosimilar experience provides a strong precedent for accepting functional equivalence despite structural differences. Regulatory agencies should establish clear guidance on SDPT classification criteria and acceptable analytical packages, reducing uncertainty for developers. International harmonization through ICH guidelines would prevent duplicative requirements across regions and enable global development strategies.

Academia plays a crucial role in advancing the scientific foundation of sequence-dominant therapeutics through systematic investigation of sequence-function relationships (Baker, 2019). Research funding agencies should recognize the importance of functional characterization and sequence-based approaches, not just structural biology. Educational programs must evolve to train the next-generation of drug developers in mechanism-based approaches, machine learning, and functional screening in addition to traditional structural biology. Academic-industry partnerships can advance precompetitive research on sequence-function relationships and develop open-source tools for the community.

Investors should recognize the value proposition of SDPT-optimized development and adjust valuation models to reflect reduced risk and accelerated timelines for qualifying programs (Maksabedian Hernandez et al., 2022). Due diligence should evaluate whether companies are applying mechanism-appropriate development strategies rather than blindly following structural approaches. Investment in platform technologies that enable sequence-based discovery and development will likely generate superior returns compared to traditional structural biology infrastructure. The transformation will require patient capital willing to support organizational changes and capability building over several years.

12.5 Final perspective

As we stand at this inflection point in the history of biological drug discovery, the choice facing the biopharmaceutical industry is clear. We can continue applying universal structure-based approaches that have yielded only limited productivity gains despite remarkable technological advances, thereby maintaining the status quo of high failure rates, lengthy development timelines, and limited global access to biological medicines. Alternatively, we can adopt mechanism-aligned strategies that align analytical and development methods with biological reality, focusing on functional validation for SDPTs while reserving structural approaches for proteins that genuinely require them.

The evidence presented in this review, drawn from biosimilar development, computational analysis, and economic modeling, strongly supports the latter approach. The recognition that sequence dominance characterizes most successful protein therapeutics is not merely an academic insight but a practical framework that can transform drug development efficiency, reduce costs, and expand patient access globally. The potential benefits—30%–40% reduction in development time, $100–250 million in cost savings per program, and a doubling of success rates—justify the organizational and cultural changes required for implementation.

The transformation will not happen overnight, and there will be resistance from those invested in traditional approaches. However, the companies, regulators, and researchers who recognize and act on these insights will shape the future of biological drug discovery. By 2035, we envision an industry in which therapeutic proteins are routinely discovered and developed in months rather than years, success rates approach 30% rather than today’s 14%, biosimilars are available globally at affordable prices. Personalized biological medicines are feasible for individual patients.

This future is not merely aspirational but achievable through the systematic application of SDPT principles, combined with advances in artificial intelligence, automation, and synthetic biology. The convergence of scientific understanding, technological capability, and regulatory evolution creates a unique opportunity to improve how we discover and develop biological medicines fundamentally. The beneficiaries will extend far beyond the companies that successfully implement these approaches to include the billions of patients worldwide who currently lack access to biological therapies.

Several limitations of the present analysis warrant acknowledgment. Because the biosimilar analysis relies on publicly available datasets and literature-derived benchmarks, and the economic projections on scenario modeling rather than empirical program-level data, the conclusions should be interpreted as conceptual and hypothesis-generating rather than definitive quantitative comparisons. The SDPT classification criteria, while operationally grounded, have not yet been prospectively validated in a controlled head-to-head development program, and the proposed regulatory frameworks require formal agency evaluation before adoption. These limitations do not diminish the conceptual value of the framework but underscore the importance of prospective studies, precompetitive data sharing, and regulatory dialogue to translate the framework into evidence-based practice. The post-structural era in biologics development is beginning. It will be characterized not by the abandonment of structural biology but by its appropriate application where mechanism demands it, while recognizing that for most therapeutic proteins, the sequence is the structure that matters. This insight, validated by extensive empirical evidence and enabled by new technologies, offers a path to more efficient, accessible, and successful therapeutic development. The framework presented in this review provides both the conceptual foundation and practical roadmap for this essential evolution. The question is not whether this transformation will occur, but how quickly the industry will embrace it and who will lead the change.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Grammar check, figure drawing.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Author disclaimer

Percentages and correlations are derived from regulatory datasets and published clinical studies and should be interpreted as approximate, mechanism-dependent estimates rather than immutable constants.

Footnotes

Edited by: Yiannis Sarigiannis, University of Nicosia, Greece

Reviewed by: Manos Christos Vlasiou, University of Nicosia, Cyprus

Androulla Miliotou, University of Nicosia, Cyprus