Quantum computing in drug discovery

1 Introduction

Pharmaceutical development proceeds through the following sequential stages: (1) target identification and biological validation; (2) hit discovery and selection; (3) lead optimisation; and (4) preclinical and clinical evaluation. Despite decades of methodological progress across experimental biology, medicinal chemistry, and high-throughput screening, the aggregate probability of a candidate entering clinical trials eventually gaining regulatory approval remains below 10%. This low success rate reflects, in large part, the extraordinary cost burden of the pipeline (exceeding USD 2 billion per approved medicine on average) and timelines spanning more than 10 years from discovery to market (Hughes et al., 2011; Mullard, 2016) and relies on computational methods with limited predictive fidelity when extrapolated beyond their training domains. Computer-aided drug design (CADD) has been adopted broadly to make the early pipeline more tractable, enabling systematic exploration of chemical space, prediction of ligand binding geometries, and multi-parameter optimisation of pharmacokinetic and ADMET profiles. Nevertheless, the classical computational engines that underpin CADD—molecular mechanics force fields and simplified semi-empirical treatments—are constructed on assumptions that deliberately sacrifice electronic accuracy for computational tractability. This trade-off limits their ability to faithfully represent the quantum mechanical phenomena—charge redistribution, polarisation, van der Waals dispersion, and electron correlation—that ultimately govern binding selectivity, enzymatic reactivity, and off-target interactions. A physically accurate, non-approximate description of drug–target interactions requires quantum-mechanical treatment of the electronic structure, particularly in systems involving transition-metal centres, reactive covalent warheads, and charge-transfer interactions at enzyme active sites. High-level wavefunction methods and hybrid DFT approaches can deliver the required accuracy, but their computational cost grows steeply—as O(N⁷) or worse for coupled-cluster methods—with the number of electrons, making them impractical for all but small model systems on classical hardware. As a result, practitioners routinely accept approximate molecular-mechanics descriptions to preserve computational feasibility, introducing systematic errors in binding affinity prediction that propagate through hit identification and lead optimisation. Overcoming this accuracy–scalability dilemma is arguably the defining unsolved problem of computational drug discovery. Quantum computation exploits the physical principles of quantum mechanics at the hardware level: qubits—unlike classical binary switches—can be maintained in coherent superpositions and mutually entangled, allowing quantum processors to manipulate exponentially large state spaces with polynomial numbers of physical operations for certain problem classes. This architectural difference underpins the prospective advantage of quantum computers for electronic structure simulation: a quantum processor can naturally encode a molecular wavefunction in a way that would require exponentially many classical bits to represent explicitly. For drug discovery, this suggests a path toward binding affinity calculations, reaction pathway analysis, and molecular property prediction at accuracy levels currently unattainable within practical classical runtimes. Progress in superconducting and trapped-ion hardware has produced NISQ processors with tens to hundreds of noisy qubits, sufficient for hybrid quantum-classical variational algorithms such as VQE to tackle toy systems (H₂, LiH, BeH₂) far smaller than any drug-like molecule. These early demonstrations show that quantum computation can contribute to chemistry without the overhead of full error correction, but it remains an open and vigorously debated question whether NISQ-scale devices will ever deliver practically useful results for drug-sized molecules (Preskill, 2018). The appropriate framing is therefore one of gradual, modular integration: quantum subroutines addressing specific electronic-structure bottlenecks within otherwise classical CADD workflows, rather than wholesale quantum replacement of the pipeline. Despite growing research output, the current body of literature on quantum computing for drug discovery remains fragmented: most studies address isolated algorithms on small benchmark molecules or provide theoretical resource estimates without connection to realistic pharmaceutical workflows. A coherent framework that maps quantum methods to each stage of the drug discovery pipeline and critically evaluates near-term versus long-term prospects is largely absent from the literature. The most closely related work is Santagati et al. (Santagati et al., 2024), which provides an excellent overview of quantum algorithms for drug design; the present article complements that work with a detailed NISQ–FTQC comparison, explicit CYP3A4 resource estimates, and a classical–quantum benchmarking table. The present article addresses this gap. Its principal contributions are: (i) a critical analysis of how quantum algorithms may reduce the accuracy–scalability trade-off in molecular modelling; (ii) a systematic mapping of quantum methods to specific drug discovery stages, from target identification to lead optimisation; (iii) a comparative framework (Table 1) benchmarking quantum against classical approaches with explicit resource estimates; and (iv) honest evaluation of current limitations, including scalability, error mitigation, and the gap between NISQ-era research and fault-tolerant quantum computing targets.

2 Overview of computational methods in drug discovery

The pharmaceutical development pipeline progresses through four broad phases: (1) target identification and characterisation, where a disease-relevant biological macromolecule is selected and validated; (2) hit discovery, where chemical compounds showing initial activity against the target are identified; (3) lead optimisation, where promising hits are refined for potency, selectivity, and drug-likeness; and (4) preclinical and clinical evaluation, where safety and efficacy are tested in animal models and human volunteers. CADD has its greatest impact in phases (1)–(3), where quantum methods also offer the most significant prospective advantages. In the later clinical stages, molecular modelling plays a minimal role, since ADMET optimisation is properly addressed during lead optimisation (phase 3), well before regulatory submission. CADD methodologies applied in hit discovery and lead optimisation divide broadly into structure-based approaches, which exploit the known three-dimensional architecture of the biological target, and ligand-based approaches, which reason from patterns in known active compounds (Sliwoski et al., 2014). Structure-based approaches require a three-dimensional protein structure obtained via X-ray crystallography, cryo-EM, or computational prediction (Zhang, 2008), and use molecular docking to identify optimal ligand conformations in the binding pocket (Kalyaanamoorthy and Chen, 2011; Anderson, 2003). Ligand-based approaches rank candidates by similarity to known actives or via QSAR models correlating activity to molecular descriptors (Golbraikh et al., 2012; Bajorath, 2015). Both families share a fundamental limitation: inability to capture polarisation, charge transfer, and electron correlation accurately. These electronic effects require quantum-mechanical treatment, motivating the quantum computing approaches discussed in Sections 4,5.

2.1 Target identification and characterization

Target identification establishes which biological entity—most often a protein (enzyme, receptor, ion channel, or transporter)—plays a causal role in a disease pathway such that its modulation by a small molecule (NCE) or biologic (NBE) produces a therapeutically useful effect. Proteins are the dominant target class because their three-dimensional binding sites are amenable to small-molecule design; nucleic acid targets represent an expanding but distinct area not covered in depth here. A practically druggable target must be capable of binding small molecules with sufficient affinity and selectivity to elicit a biologically significant response both in vitro and in vivo. While all proteins are theoretically tractable, practical druggability is constrained by binding site geometry, conformational flexibility, and current technological limitations — as exemplified by historically challenging targets such as KRAS, which required multi-mechanism approaches (Hughes et al., 2011). Precise target identification and validation greatly increases the probability of success in subsequent discovery steps as well as aids in predicting side effects of target modulation (Hughes et al., 2011).

Classical target identification tools include affinity chromatography (Lomenick et al., 2010; Burdine and Kodadek, 2004; Chan et al., 2010; Schenone et al., 2013), genetic screening, gene expression profiling, and systems biology network analysis (Czodrowski et al., 2009; Yang et al., 2012; Katsila et al., 2016; Nidhi et al., 2006). Each approach has limitations in specificity or throughput. Quantum computing could provide solutions at this stage because the bottleneck is the dimensionality and noise of multi-omics datasets: classical machine learning struggles with high-dimensional genomic feature spaces where sample sizes are small relative to feature count. Quantum-enhanced kernel methods and QML classifiers could in principle better exploit such data regimes for target prioritisation, as discussed in Section 7.2.

Genetic screening, gene expression profiling, and systems biology network analysis complement biochemical target identification, allowing causal disease genes and regulatory pathway nodes to be identified from omics datasets (Schenone et al., 2013; Zheng et al., 2004; Czodrowski et al., 2009; Yang et al., 2012; Katsila et al., 2016; Nidhi et al., 2006). Machine learning models including Bayesian classifiers assist in target prioritisation across databases and literature (Yang et al., 2012; Katsila et al., 2016; Nidhi et al., 2006). The quantum relevance of these approaches is discussed in Section 7.2.

For structure-based drug development, structural characterisation of the target protein is crucial after target identification. Protein purification and crystallization difficulties frequently limit the utilization of experimental methods like X-ray crystallography and NMR spectroscopy (Grey and Thompson, 2010). When no experimental structure is available, computational tools such as threading, comparison modelling and ab initio structure prediction are employed. Comparative modeling is the most common method to build 3-D protein structures by utilizing empirical structural models having known sequence similarities in the protein Data Bank (PDB) (Zhang, 2008) and (Bernstein et al., 1977). Regions where the structures do not have suitable templates are refined using molecular dynamics simulations, Monte Carlo minimization or genetic algorithms (Sliwoski et al., 2014).

When appropriate templates are not available, free modeling methods can be used, including knowledge-based fragment assembly and physically justified methods like molecular dynamics simulations. Physical-based methods can offer information about the routes of protein folding, but are numerically intensive and require accurate force fields and extensive conformational sampling (Zhang, 2008; Bradley et al., 2005). The difficulties underline the capability of quantum simulation and optimization techniques, that can prove helpful in solving complex issues such as protein-folding. The identification of ligand binding and therapeutic pockets of the protein is called target characterization (Kalyaanamoorthy and Chen, 2011). It can be used directly if the binding site is known, as in the case of a co-crystallized complex of protein and ligand. If the binding site is not known, computational tools like Ligsite, Qsite Finder and CASTp are used to predict the likely binding site by using geometry and energy considerations (Laurie et al., 2006; Henri et al., 2010). The computational tools such as QSiteFinder and CASTp predict binding pockets from geometry and electrostatics. The bottleneck here is accuracy for flexible or novel targets; quantum simulation of protein electronic structure could in principle improve pocket identification where classical force fields fail (Section 5).

2.2 Hit search and lead identification

Hit identification finding compounds with measurable target activity has shifted from costly physical high-throughput screening toward virtual screening, which computationally prioritises candidates using either binding-site structure or statistical models derived from known actives, before any compound is synthesised.

Ligand-based screening ranks candidates by similarity to known actives or via QSAR models mapping activity to molecular descriptors including quantum-mechanically derived charge and orbital representations (Acharya et al., 2011; Marrero-Ponce et al., 2012; Auer and Bajorath, 2008; Bajorath, 2002; Ekins et al., 2007; Zhang, 2011). Both approaches degrade sharply outside the chemical domain of the training data and offer no mechanistic rationale for activity trends, a gap that electronic-structure descriptors computed via quantum chemistry could in principle narrow. Early linear QSAR models (MLR, PCA, PLS) have largely been superseded by machine learning methods (ANN, SVM, random forests) that better capture the nonlinear structure–activity relationship (Xiang et al., 2012; Zhang et al., 2017; Colwell, 2018; Lima et al., 2016; Chen et al., 2018; Popova et al., 2018; Gomez-Bombarelli et al., 2018; Luechtefeld et al., 2018).

Structure-based hit search instead predicts ligand–protein binding via molecular docking and scoring (Sliwoski et al., 2014; Halperin et al., 2002; Connolly, 1983; Ryde and Soderhjelm, 2016; Halgren, 1996; OBoyle et al., 2009). Its principal limitation is well established: standard molecular-mechanics (MM) scoring functions cannot fully capture polarisation and charge transfer, which are decisive in enzyme catalysis and metalloprotein binding (Ryde and Soderhjelm, 2016; van der Vaart and Merz, 1999; van der Vaart and Merz, 2000; Friesner, 2005; Bryce, 2011; Raha et al., 2007). QM/MM hybrid scoring addresses this by treating the ligand and key residues quantum-mechanically, but its computational cost confines it to the late-stage lead optimisation funnel rather than large-scale screening (Ryde and Soderhjelm, 2016; Rao et al., 2013; Zhou and Caflisch, 2010). This cost barrier — not a lack of theoretical accuracy is precisely the bottleneck that quantum-hardware acceleration of the QM region could address, in principle (Section 5).

2.3 Lead discovery and optimization

After hit compounds are identified, they undergo a lead discovery and optimization phase aimed at generating a smaller set of candidates with improved drug-like properties. In order to improve pharmacokinetic and ADMET profiles in addition to biological activity, this step entails iterative cycles of CADD-guided design, chemical synthesis, and in vitro and in vivo testing (Sliwoski et al., 2014). Computational guidelines are particularly useful since it is assumed that successive chemical transformations will lead to ongoing changes in the properties of medications. Because they enable the rapid assessment of structural changes and the determination of the influence of structural modifications on activity and drug-likeness, particularly when comprehensive target structural data is lacking, QSAR models constructed with smaller, more targeted datasets are an important component of lead optimization. However, as optimization continues, structure-based approaches become more important, since the correct prediction of binding affinity is an important intermediate step that allows lead candidates to be prioritized (Raha et al., 2007). The effect of even small binding energy differences on the binding equilibrium constants is magnified orders of magnitude and thus has a significant impact on biological efficacy.

Most binding affinity predictions methods rely on molecular mechanics (MM) force fields, which are not designed to capture important quantum mechanical effects such as polarisation, charge transfer, and electron correlation. These effects are particularly important in reactive binding processes, metalloproteins, and enzyme catalysis. Although these phenomena are naturally incorporated in quantum mechanical (QM) approaches, they are not commonly exploited in drug development because they are computationally impractical on conventional computers, particularly in large ligand-protein complexes. Finding more efficient techniques for quantum simulation is crucial since approximation QM approaches are typically restricted to tiny subsystems of the system and can contain significant mistakes depending on the level of theory used.

Existing QM-based binding-affinity methods fall into three tiers of increasing rigour and cost: single-structure calculations on a fixed MM-optimised geometry (cheap but inaccurate); end-point methods that sample multiple conformations with a partial QM/MM treatment (intermediate cost and accuracy); and full free-energy simulations (FES/FEP) that exhaustively sample the binding pathway at high precision but at prohibitive computational cost for routine use (Ryde and Soderhjelm, 2016; Zhang and Zhang, 2003; Fedorov and Kitaura; Cavasotto et al., 2018; Luzhkov and Warshel, 1992; Rod and Ryde, 2005; Duarte et al., 2015; Abel et al., 2017). The progression from semiempirical methods through DFT to wavefunction-based CCSD(T) tracks steadily improving accuracy at steeply increasing cost (Ryde and Soderhjelm, 2016). This cost–accuracy ceiling is the key computational limitation that a quantum algorithm capable of efficiently simulating the QM active site could, in principle, overcome consistent with the language convention stated above.

3 Quantum computing

Quantum computation encodes information in qubits rather than classical bits. A qubit can occupy a coherent superposition of |0⟩ and |1⟩, and n entangled qubits span a 2ⁿ-dimensional Hilbert space that grows exponentially with n. Quantum algorithms exploit this structure for provable speedups: Shor’s algorithm factors integers in polynomial time, Grover’s search achieves quadratic speedup, and quantum simulation algorithms encode molecular Hamiltonians into qubit operators in a way that mirrors the physical state space naturally.

Multiple physical hardware platforms have reached the NISQ regime, including superconducting transmon circuits, trapped ytterbium and barium ions, neutral Rydberg atom arrays, and photonic processors. In fact, proof-of-principle experiments performed on these platforms have shown that existing devices are capable of creating quantum states that are classically intractable (Preskill, 2018; Aaronson and Chen, 2016; Boixo et al., 2018) and thus demonstrate the principle of quantum computational advantage. However, all current processors remain vulnerable to gate errors, qubit decoherence, and readout infidelity that accumulate with circuit depth. Full quantum error correction via the surface code or related topological schemes can suppress logical error rates below any target threshold, but the required ratio of physical-to-logical qubits estimated at roughly 1,000:1 at current physical error rates places genuinely fault-tolerant operation well beyond present hardware.

NISQ processors have been deployed within hybrid quantum-classical frameworks for variational algorithms, of which VQE is the primary example (McClean et al., 2016). A parameterised quantum circuit prepares a trial wavefunction; the quantum processor measures the energy expectation value; and a classical optimiser adjusts the circuit parameters iteratively. It must be emphasised, however, that no NISQ-based hybrid workflow has yet solved a chemically or pharmaceutically industrially relevant problem — all published demonstrations remain on toy systems (H₂, LiH, BeH₂) with active spaces of 12–16 orbitals (mapped to 12–16 qubits via Jordan–Wigner encoding) (Preskill, 2018; McClean et al., 2016; Goings et al., 2022). Whether NISQ devices can scale beyond these toy systems to drug-relevant electronic structure problems at chemical accuracy is an open and actively debated question (Preskill, 2018; Bharti et al., 2022). The hybrid framework is therefore best understood as a research direction: it has not been deployed in any industrial pharmaceutical workflow, no drug-relevant problem has been solved using it, and all published demonstrations remain on toy systems far smaller than any drug candidate. It is a promising methodological approach that requires significant hardware and algorithmic advances before practical deployment becomes feasible.

4 Quantum computing techniques for drug discovery

4.1 Quantum algorithms for drug and compound design

Quantum algorithms address the fundamental computational bottleneck of classical chemistry methods: the exponential scaling of exact methods such as Full Configuration Interaction (FCI), whose cost grows as O(N!) with system size, making them intractable for molecules beyond a few tens of electrons on classical hardware (Montanaro, 2016). It is essential to distinguish between NISQ-compatible near-term algorithms and those requiring fault-tolerant quantum computing (FTQC). The quantum chemistry methods underpinning these algorithms—including DFT, ab initio wavefunction theory, and VQE—are treated in detail in Section 5. Grover’s search algorithm (Grover, 1996) achieves a quadratic speedup for searching unsorted databases and is conceptually relevant for screening chemical libraries against molecular specifications. But practical implementations will rely on the Quantum Phase Estimation (QPE) subroutine, which will require fully error corrected qubits, which are not present in existing NISQ devices. Grover’s algorithm is thus not a near-term NISQ application, but rather an FTQC target that requires fully error-corrected logical qubits. The most widely used algorithms from the NISQ era are the Variational Quantum Eigensolver (VQE) which is used for approximating the ground state energy of a molecular Hamiltonian with short-depth parameterized circuits optimized classically (Peruzzo et al., 2014). VQE is especially well adapted to the calculation of the lowest electronic states of molecules that are key to prediction of binding geometry, reactivity, and pharmacological activity. As molecular complexity grows, classical computers find it hard to model it, while VQE is able to model electronic wavefunctions more realistically, and, perhaps, in a scalable way that can simulate drug-relevant molecular interactions (Singh et al., 2024). Only a few small molecules (H₂, LiH, BeH₂) have been demonstrated so far. Critically, there is no guarantee that VQE will scale to drug-sized systems: the variational optimisation landscape is known to exhibit exponentially flat ‘barren plateaus’ as system size grows, making gradient-based optimisation increasingly difficult. VQE is fundamentally a heuristic Ritz-type algorithm — convergence to the true ground state is not guaranteed, and the choice of ansatz strongly determines both accuracy and trainability. Whether any NISQ-compatible variational approach can deliver chemical accuracy (<1 kcal/mol) for drug-relevant active spaces remains an open research question (Preskill, 2018; Bharti et al., 2022; Elfving et al., 2020).

4.2 QML in drug discovery: potential and limitations

Quantum machine learning (QML) explores whether quantum computational resources can augment statistical learning for drug property prediction and diagnostics (Siddique and Chow, 2021). Quantum-enhanced kernel methods including QSVM implicitly compute inner products in large Hilbert space feature maps via quantum circuits. However, critical limitations must be acknowledged. First, current benchmarks consistently show classical deep learning (graph neural networks, transformers) outperforming demonstrated QML approaches on standard drug discovery benchmarks (Sajjan et al., 2022). Second, the ‘quantum RAM’ problem efficiently loading large classical datasets onto a quantum processor remains unsolved and may offset any computational advantage for general QML. One partial solution under active investigation is the quantum streaming model, which processes data sequentially rather than requiring full quantum memory loading; recent work from Google and collaborators (Google Quantum AI et al., 2025) demonstrates that streaming quantum algorithms can bypass some quantum RAM requirements for specific learning tasks, though applicability to drug-relevant molecular datasets remains to be established. Third, whether quantum-mechanical descriptors or wavefunctions could genuinely enrich ML training data for molecular property prediction is largely unexplored (Wei et al., 2023; Goyal, 2022; Prajapati et al., 2023). QML is therefore a research-stage concept with genuine theoretical motivation but no demonstrated practical advantage in drug discovery to date. For a comprehensive recent assessment of QML capabilities and limitations relevant to life sciences, (Schuld and Petruccione, 2026).

4.3 Quantum sensing for biomedical applications

Quantum sensors exploiting entanglement and coherence have demonstrated enhanced sensitivity to weak magnetic fields, with relevance to MRI and magnetoencephalography instrumentation (Aslam et al., 2023). Quantum sensing is also being investigated for molecular detection and single-molecule biosensing. These approaches remain pre-clinical but represent an active Frontier distinct from the quantum computing methods discussed elsewhere in this review.

4.4 Quantum-optimized treatment planning

Radiotherapy treatment planning requires finding optimal beam angles and dose distributions to maximise tumour control while minimising damage to surrounding healthy tissues (Chen and Kung, 2017). The classical bottleneck is combinatorial: the search space of possible beam configurations grows exponentially with the number of beams and dose levels, and current clinical solvers use heuristic approximations that do not guarantee global optimality. Quantum optimisation algorithms — specifically QAOA — could in principle search this combinatorial space more efficiently than classical heuristics (Nachman et al., 2021), because QAOA leverages quantum superposition to evaluate multiple beam configurations simultaneously. However, current NISQ hardware cannot handle real clinical problem sizes, so this remains a longer-term target beyond the primary drug discovery scope of this review.

5 Quantum chemistry for drug design

Note: Section 4.1 introduced VQE as the primary near-term quantum algorithm for electronic structure. This section describes the classical quantum chemistry methods DFT, ab initio wavefunction approaches, and their drug-design applications that represent the state-of-the-art baseline which quantum algorithms aim to augment or replace. The rational design of a drug depends on accurate knowledge of the electronic structure of a molecule because all properties such as binding affinity, selectivity, metabolic stability, and off-target liability are rooted in the distribution of electrons and the electrostatic fields they generate (Zhou et al., 2010). Quantum chemistry provides first-principles tools to compute molecular properties, most valuably where force-field approximations are known to fail: transition-metal active sites, covalent inhibitors, and strong charge-transfer interactions (Manathunga et al., 2022). It is important to note that many clinically successful drugs have been designed using force-field or empirical methods; quantum chemistry is not universally required, but is selectively justified where electronic effects are mechanistically decisive. The practical challenge is achieving this accuracy for drug-relevant system sizes and conformational ensembles within tractable computational budgets — which is precisely where quantum computing offers a prospective path forward.

• High-Throughput Screening: To rapidly evaluate sizable libraries of possible drug candidates and rank compounds for experimental testing, high-throughput virtual screening can employ quantum chemistry computations (Irfan et al., 2022).
• Electronic Structure Calculation: Quantum chemical methodologies including Hartree–Fock, correlated wavefunction methods, and Density Functional Theory (DFT) are used to compute the electronic structure of molecules (Bowler and Miyazaki, 2012). It is important to note that all of these methods are approximations: Hartree–Fock neglects electron correlation; DFT depends on an unknown exact exchange-correlation functional and uses approximations; even the highest-level coupled-cluster methods truncate the cluster expansion. No classical method solves the Schrödinger equation exactly for many-electron systems — this is precisely the gap that quantum computers aim to address. The electron density distribution and energy levels computed by these approximate methods are central to understanding molecular reactivity and physicochemical properties.
• Binding Energy and Affinity: Quantum chemical methods can compute the interaction energy between a drug candidate and its target protein by solving the electronic Schrödinger equation for the active site (Guareschi et al., 2023). Reliable binding affinity prediction requires sampling an ensemble of conformations rather than a single optimised geometry: the thermodynamic binding free energy reflects contributions from multiple conformers, protonation states, tautomers, solvent interactions, and protein flexibility. A quantum-computed single-geometry energy provides a rigorous electronic-structure reference for the active site, which is most valuable when combined with classical conformational sampling methods such as molecular dynamics or FEP to yield complete binding affinity estimates (Kalyaanamoorthy and Chen, 2011; Ryde and Soderhjelm, 2016; Guareschi et al., 2023).
• Transition State Analysis: In order to examine the chemical changes that occur during medication processing and metabolism, we must comprehend the reaction mechanism and transition states, which are supplied by quantum chemistry (Schramm, 2013).
• Solvent Effects: Drugs’ solubility, stability, and bioavailability cannot be predicted without understanding how they behave in various solvents (Reichardt, 2007). Quantum chemistry can help explain how solvents affect molecular interactions.
• Electrostatic Interactions and Charge Distribution: Researchers must comprehend how electrostatic interactions impact binding and reactivity, and quantum chemistry helps shed light on this charge distribution (Sebens, 2021).
• Prediction of Spectroscopic Properties: Quantum chemistry techniques can be used to predict spectroscopic characteristics like vibrational frequencies, NMR chemical shifts, and UV-visible absorption spectra (Barone et al., 2021). These predictions may be used for molecular characterization and for interpretation of experimental spectroscopic data.

5.1 Quantum computational methods for molecular structure prediction

The use of quantum computational methods for the prediction of molecular structure relies on quantum mechanics to predict the arrangement of atoms and the distribution of electrons of molecules to calculate their reactivity and the reactivity of drug–target interactions (McDouall, 2013). While classical force-field approaches rely on parametrised potentials for the interactions between point charges, quantum approaches explicitly solve the electronic Schrödinger equation and provide a physically realistic description of bond formation, charge redistribution, and dispersion interactions that cannot be captured by classical approaches. Presently, the technologies that are under investigation to predict the molecular structures as follows.

5.1.1 Density functional theory (DFT)

Density Functional Theory (DFT) is a classical computational method grounded in quantum mechanics that reformulates the many-electron problem in terms of electron density rather than the full wavefunction, making it tractable for systems of hundreds of atoms on conventional hardware (Orio et al., 2009; Engel and Dreizler, 2011). It is important to clarify that DFT is not a quantum computing method it runs on classical CPUs and GPUs. The Hohenberg–Kohn theorems provide the theoretical foundation; the Kohn–Sham equations enable practical implementation. The accuracy of DFT depends on the choice of exchange-correlation functional: hybrid functionals such as B3LYP and ωB97X-D provide reliable thermochemistry for many drug-relevant systems but fail for strongly correlated systems including metalloprotein active sites and open-shell transition-metal complexes. The relevance of DFT to quantum computing is that quantum processors could in principle compute the exchange-correlation energy exactly, providing accuracy beyond any classical DFT functional for precisely these difficult cases.

5.1.2 AB initio methods

In addition to DFT, there is also a systematic hierarchy of accuracy in correlated wavefunction methods. Methods that follow the HF level incorporate increasing amounts of instantaneous electron–electron correlation as the approximate mean-field approximation, which is recovered at a higher cost of computing power: Møller–Plesset perturbation theory (MP2), coupled-cluster with singles, doubles and perturbative triples [CCSD(T)], and complete active space self-consistent field (CASSCF) (Levine et al., 2009). Although CCSD(T) is regarded as the ultimate standard for reaction energetics and non-covalent interaction energies, its O(N⁷) scaling prevents its use for molecules with 20–30 atoms, which are typical in drug applications, without quantum acceleration. Adopting quantum computing techniques for molecular structure prediction would involve researchers in quantum mechanics, the ability to use existing programs for computational chemistry, and the knowledge of using algorithms and methods that are appropriate to the research goals (Ahmed et al., 2013). With the advancement of the field, these methods could provide researchers with better and more detailed understanding of molecular behaviour, leading to innovations and discoveries.

There is great potential for added value to molecular structure prediction from quantum computing when combined with both ab initio methods and density functional theory (DFT). Such classical methods for the simulation of the electronic structure of molecules are powerful, but have a poor computational scaling with system size and are not applicable for large or complex molecular systems. For some of these problems, quantum computers can run certain quantum simulations more efficiently, providing a significant computational advantage for these problems.

The latter attribute is especially helpful in the analysis of large molecules or complex chemical reactions. The computational feasibility may be improved by providing fast solution to the electronic structural problems that are difficult to solve with traditional ab initio methods, using quantum computers. Furthermore, the more accurate the quantum computers can be in simulating the interactions between electrons and the quantum events, the more accurate the quantum computers can be. Such greater accuracy could result in more accurate predictions of molecular properties and dynamics, such as spectroscopic characteristics, bond dissociation energies and reaction pathways. However, a key caveat is that all the quantum chemistry calculations, including those performed on classical computers as well as those to be performed on quantum computers, are currently based on a single optimized molecular geometry. The thermodynamic and kinetic properties of a molecule in solution are determined by an ensemble of conformers, tautomers and protonation states, all of which are important for drug-target binding affinity and selectivity. These factors of solvent interactions, protein flexibility and conformational entropy play a crucial role in determining the ability of a designed molecule to bind effectively in a biological context (Kalyaanamoorthy and Chen, 2011; Anderson, 2003; Ryde and Soderhjelm, 2016). Quantum simulation of a single geometry, while informative, is thus only one component of a complete binding affinity prediction. Future quantum computing approaches for drug discovery must ultimately address conformational sampling, for example, by combining quantum electronic structure calculations with classical or quantum-enhanced molecular dynamics simulations to achieve predictions that are meaningful in real-world drug discovery pipelines.

5.2 Potential energy surfaces and reaction pathways

Potential energy surfaces (PES) describe how the total energy of a molecular system varies with atomic positions, providing a theoretical map of equilibrium geometries, reaction pathways, and transition states (Teplukhin and Babikov, 2015; Schatz, 1989). Quantum computing offers the prospect of computing PES with higher accuracy than classical DFT for systems where electron correlation is critical, such as metalloprotein active sites and reactive intermediates. By using QC’s computer powers to improve PES calculations, drug development might become more accurate (Baiardi et al., 2023). In order to compute reaction rates and procedures, it also aids in locating transition states on PES. This quantum-precise control is applied in environmental chemistry, nanomaterials and materials science, and catalysis to speed up the optimization of drugs, development of synthetic routes, and study of complex chemical reactions. A key component of contemporary drug discovery is quality control. In order to explore possible energy surfaces and reaction pathways, the research goals should be carried out using quantum chemical methodologies, computer algorithms, and visualization tools (Baiardi et al., 2021). Scientists examine the specifics of molecular dynamics and energetics to understand the fundamental mechanisms behind chemical reactions. These findings have significant ramifications for drug development, material science, environmental chemistry, and catalysis.

6 Complete pipeline for drug development using quantum computing

The drug development pipeline is a methodical procedure that starts with quantum algorithms and simulations of molecular-level explanations of intricate biomolecular systems and disease processes in order to identify and characterize targets (Amaro and Mulholland, 2018). The quantum-enhanced virtual screening helps to identify potential drug candidates in large chemical libraries more quickly, and is the starting point in subsequent steps, such as hit identification (Bonde et al., 2024).

In the lead optimisation stage, quantum simulations could in principle guide the prediction of binding affinity and ADMET properties for a focused set of candidates (Richings and Habershon, 2022). The specific quantum contribution here is the VQE-based active-site energy evaluation (as detailed in the CYP3A4 workflow, Section 8.5), which could replace the approximate QM/MM treatment currently used for metalloprotein targets (Schaduangrat et al., 2020; Liu et al., 2024). QML-based QSAR models may also assist in structure–activity relationship prediction for ligand-based lead optimisation (Niazi and Mariam, 2023).

To summarise the pipeline mapping: NISQ-era VQE is the primary near-term tool for small active-space electronic structure calculations; QPE on FTQC hardware is the long-term target for full-active-space accuracy on drug-sized molecules; and QML classifiers are a research-stage tool for property and activity prediction. Each technology has distinct hardware requirements, timelines, and accuracy limitations, as summarised in Table 1 and the CYP3A4 resource estimates in Section 8.5. Claims of transformative impact must be qualified by the current state of hardware maturity (Hassanzadeh, 2020; Sajjan et al., 2022).

Early applications of quantum computing are now emerging in the drug discovery process, but it is worth noting that most of these applications are still proof-of-concept, and do not yet represent truly quantum pipelines (Outeiral et al., 2021). Partnerships are growing: Boehringer Ingelheim and Google have collaborated on quantum computing for drug discovery (Ingelheim, 2021), while Merck and Zapata Computing investigated quantum-enhanced simulation of protein–ligand interactions (Streif et al., 2023) and Merck and Zapata Computing have explored the use of quantum computing to simulate protein-ligand interactions (KGaA, 2021; Casares et al., 2022). Currently, quantum simulation methods to design antibiotics are still being explored by Vertex Pharmaceuticals and other entities (Bernal et al., 2023; McArdle et al., 2020). Note that quantitative statements from references (Ingelheim, 2021; Streif et al., 2023; KGaA, 2021; Casares et al., 2022; Bernal et al., 2023; McArdle et al., 2020) are based on company statements, and not peer reviewed publications, and that the independent verification of such performance figures is yet to come. As the field continues to develop, comparing the performance of the field to state-of-the-art classical approaches will be key to achieving true quantum advantage.

Besides these, the life sciences are also becoming a significant area for quantum technologies due to their use in drug development, simulation of chemical processes and genomic analysis (Gunashekar et al., 2022). Despite the fact that the field is at its infantile phase, these advancements highlight the increasing and real presence of quantum computing in drug discovery. As the technology matures, the impact of the technology is expected to be further enhanced, leading to more efficient pipelines and useful therapeutic products.

7 Broader healthcare applications of quantum computing

7.1 Quantum-accelerated molecular modeling in pharmaceutical research

One of the fundamental challenges of molecular modelling in pharmaceutical research is the trade-off between accuracy and tractability of computations, for which quantum computers are ideally suited (Cao et al., 2019). A mechanistically approximate classical force-field simulation is fast for a protein–ligand complex, whereas an accurate study of electronic effects can be performed at high level quantum chemistry, but is limited to small model systems on classical computing systems. In principle, quantum processors can fill this gap by enabling accurate electronic-structure calculations for pharmacologically relevant systems, with the reaction fidelity required to obtain reliable predictions of binding affinity and reactivity, and without the exponential computational cost of classical computers. This outlook might significantly shorten the hit to lead and lead optimisation process. Industry partnerships such as those explored by Biogen and Accenture Labs for neurological disease applications (ALS, Parkinson’s, Alzheimer’s) (Zinner et al., 2021) demonstrate growing commercial interest, though peer-reviewed benchmarked results quantifying quantum advantage over classical methods have not yet been published for any of these collaborations. Similarly, Moderna and IBM have collaborated to investigate QC applications to mRNA design relevant to vaccines (Zinner et al., 2021). These industry partnerships demonstrate growing commercial interest in quantum-accelerated drug discovery, though published, peer-reviewed benchmarks with quantified performance improvements against classical state-of-the-art remain limited at this stage.

7.2 Quantum computing and genomics for drug target discovery

Genomics is directly relevant to drug target discovery: identifying which genetic variants are causally linked to disease pathways guides selection of therapeutic targets and stratification of patient populations for clinical trials. Quantum computing offers potential advantages here through quantum-enhanced clustering and classification of high-dimensional genomic datasets (Gepp and Stocks, 2009). The Q4Bio project has explored the use of quantum algorithms to analyze DNA sequence diversity at scales that challenge classical hardware (University of Cambridge, 2024). More directly drug-discovery-relevant is the use of quantum machine learning for multi-omics data integration combining genomic, proteomic, and metabolomic profiles to identify novel disease-associated targets and biomarkers. While purely classical deep learning currently leads in benchmarks for these tasks, quantum-enhanced kernel methods may offer advantages in data regimes where sample sizes are small relative to feature dimensionality (Sajjan et al., 2022).

7.3 Cloud-based quantum platforms: access and reproducibility for quantum chemistry in drug discovery

A practical consideration specific to quantum chemistry applications in drug discovery distinct from the genomics applications in Section 7.2 is access to quantum hardware for running VQE and related molecular simulation algorithms. Cloud-based quantum computing platforms including IBM Quantum, Amazon Braket, Microsoft Azure Quantum, and Google Quantum AI now provide remote programmatic access to NISQ devices and quantum simulators. These platforms lower the barrier to entry for drug discovery researchers, enabling quantum algorithm prototyping and benchmarking without the need for on-site quantum hardware (How and Cheah, 2023). IBM Quantum Experience, for example, provides access to superconducting qubit processors with up to 127 qubits and has been used for proof-of-concept quantum chemistry calculations. With respect to pharmaceutical companies, cloud quantum platforms enable hybrid quantum-classical workflows, in which quantum subroutines (e.g., a VQE energy evaluation) are invoked from within classical molecular simulation workflows. Another issue is the repeatability of quantum calculations on the cloud hardware: These may differ significantly between devices, and from one run to the next, owing to noise from the hardware, drift in the calibrations, and queuing delays. There is a need to develop a “benchmarking” norm for reporting and reproducing results in quantum chemistry, similar to the one used in classical computational chemistry, in order to enable credible use of quantum computing in drug discovery (Elfving et al., 2020).

7.4 Regulatory and reproducibility considerations

With the addition of quantum computing techniques into the pharmaceutical workflow, regulatory and reproducibility concerns become more salient. Computational methods that are used in Drug Development submissions must be scientifically sound, well defined and reproducible in accordance with regulatory agencies like the FDA or EMA. Several challenges must be met to enable quantum methods to meet these requirements. First, quantum measurements are stochastic and hardware dependent, which makes the outcome of measurement based algorithms not reproducible on other devices even for the same algorithm. Second, unlike in classical FEP or DFT, there are no accepted community benchmarks in the field of quantum chemistry relevant to drug applications that can be used for assessing the accuracy limits needed for regulatory submission. Third, the traceability and provenance of quantum computations need to be recorded, in compliance with good manufacturing practice (GMP) and good laboratory practice (GLP). Addressing these concerns will need a concerted effort between the quantum computing, computational chemistry and pharmaceutical regulatory communities (Elfving et al., 2020). Technology Readiness Level (TRL) assessment provides a useful framework: most quantum chemistry applications for drug discovery currently sit at TRL 2–3 (technology concept formulated; experimental proof of concept), with near-term NISQ applications potentially reaching TRL 4–5 (validated in laboratory/relevant environment) within the next five to 10 years, and FTQC-based approaches remaining at TRL 1–2 for the foreseeable future.

8 Future role of quantum computation in drug discovery

8.1 Extending the applicability of computational models in chemical space

One of the main concerns of researchers exploring quantum computing has been figuring out when methods may outperform conventional hardware. For many years, even the most basic (hydrogen chains) and the more complex (FeMo cofactor nitrogenase), quantum benchmark systems had little effect on the development of medicines (Arute et al., 2020; Reiher et al., 2017; Jahangiri et al., 2024). Recently, Goings et al. applied the more useful model of cytochrome P450 enzymes (CYPs) that are critical for steroid biosynthesis and drug metabolism, but are also metalloenzymes containing heme (Goings et al., 2022). The proper identification of relevant properties of iron-containing solutions is a major challenge in computational chemistry. In the case of more complicated theoretical methods both on conventional and quantum devices this means determining the set of orbitals (the so-called active space) that is suitable to describe the physics of the system. Important, however, is the fact that the number of orbitals must be small for the conventional methods and it is not known what the computational cost will be for the same number of active orbitals with the quantum methods. Goings et al. (Goings et al., 2022) use classical approaches to estimate the classical resources required and to assess the active space size required for providing insightful chemical information about the system. By comparing this with quantum phase estimation using quantum resource estimates, a next-generation quantum approach that makes it possible to calculate energy in quantum chemistry, they demonstrated a crossover at around 50 orbitals when quantum computing might become more beneficial. The basic physics of a complex binding heme site with about 40 atoms was accurately depicted by these models.

8.2 Beyond basic ground state energy calculations

The majority of research in quantum computer chemistry to far has been on techniques for figuring out the electronic ground-state energy of different systems. But there is often more to computational chemistry in drug discovery than merely figuring out the overall energy. Cortes et al. (Cortes et al., 2024) Examine how symmetry-adapted perturbation theory (SAPT) may be applied to quantum devices. Because it makes it possible to describe non-covalent interactions in chemical systems, this method is widely utilized to enhance understanding of the binding properties of drugs in enzyme systems. This method has shown interactions on classical computers that would have been missed without quantum physics (Parrish et al., 2017; McConnell , 2021). For a heme and artemisinin complex, Cortes et al. estimated quantum resources and discussed bottlenecks in their method. Their research shows that in order to effectively improve quantum computing for drug discovery, more sophisticated algorithms that go beyond basic total-energy calculations are both possible and required.

8.3 Quantum fingerprints provide insights regarding covalent inhibitors

Covalent inhibitors are medications that attach themselves to particular proteins in an irreversible manner by forming a covalent link between the proteins and the drug’s so-called warhead, also known as the reactive site. Rather than being the product of logical design, almost all of the recent breakthroughs in covalent inhibitors have been the consequence of chance. The chemistry that involves the formation of the covalent bond between the enzyme and warhead is not known, especially how the chemical reaction that forms the bond to the warhead affects the covalent bond between the enzyme and the warhead. Although computational rational design of covalent inhibitors has become a big challenge in drug development, the computational methods are not as advanced as the rest of small molecule drug design. Montgomery et al. (Montgomery et al., 2024) tackled the problem of predicting the properties of covalent warheads using techniques from quantum computing. They use techniques from classical simulations, such as density matrix embedding theory, in addition to quantum algorithms, to build over time a specific quantum state of the weapons. They employ techniques from classical simulations, namely density matrix embedding theory, in combination with quantum algorithms to temporally construct a particular quantum state of the weapons. From the quantum properties of the time evolved states, they may be able to derive the energies of the highest occupied and lowest vacant molecular orbitals, based on DFT calculations. They point to the necessity of identifying “quantum fingerprints” that may be relevant, and used both as a way to understand the system and as a way to feed to machine learning techniques to assess a drug discovery system. Thanks to these quantum fingerprints, the differences between intrinsic and pocket-specific reactivities, i.e., the changes of covalent bond formation of covalent inhibitors that can be provoked by the environment, should be easier to understand.

A critical question for the field is: under what conditions do quantum methods offer a tangible advantage over the best available classical approaches? Table 1 summarises the comparison across the four primary computational tasks in drug discovery. For electronic structure calculations, classical DFT with hybrid functionals achieves approximately 1–3 kcal/mol accuracy for small molecules but scales as O(N3), becoming impractical beyond ∼100–200 atoms. CCSD(T) achieves sub-kcal/mol accuracy but scales as O(N7), restricting use to ∼20–30 atoms. Quantum Phase Estimation (QPE) on a fault-tolerant device could match CCSD(T) accuracy for 50+ atom systems with polynomial scaling, a genuine future advantage requiring millions of logical qubits (Goings et al., 2022; Cortes et al., 2024). For binding affinity prediction, state-of-the-art classical FEP+ methods achieve ∼1 kcal/mol accuracy for congeneric series with GPU-accelerated MD. Current NISQ VQE approaches cannot yet compete with production FEP+ for lead optimization; the prospective quantum advantage lies in treating metalloprotein active sites and covalent inhibitors at higher QM levels (Goings et al., 2022; Montgomery et al., 2024). For virtual screening, classical docking tools screen millions of compounds per hour; Grover-based quantum search offers quadratic speedup but requires fault-tolerant hardware and a structured oracle. For QML, classical deep learning currently outperforms demonstrated QML on standard drug discovery benchmarks (Siddique and Chow, 2021; Sajjan et al., 2022).

8.4 Quantum vs. classical methods for drug discovery

Table 1 compares the five main tasks in drug discovery across quantum and classical computational methods. The table shows the classical tool that is most appropriate for each task, its accuracy and major limitation, the quantum approach, the potential quantum gain, and a realistic deployment time horizon. This framework enables readers to know where quantum methods have theoretically motivated benefits, and what time scale these benefits might become practically attainable.

TABLE 1. Comparative assessment.

NISQ, Noisy Intermediate-Scale Quantum; FTQC, Fault-Tolerant Quantum Computing; QPE, quantum phase estimation; VQE, variational quantum eigensolver; QAOA, quantum approximate optimisation algorithm; GNN, graph neural network; QSVM, quantum support vector machine; QNN, quantum neural network; FEP+ = free energy perturbation plus; DFT, density functional theory; CCSD(T) = coupled cluster singles, Doubles, and perturbative Triples.

8.5 Concrete hybrid quantum-classical workflow with resource estimates

To concretely illustrate where quantum computing fits into an existing drug discovery pipeline, we present a worked hybrid workflow using cytochrome P450 3A4 (CYP3A4) as the target system. CYP3A4 is a clinically critical metalloenzyme responsible for metabolising approximately 50% of marketed drugs; its iron-containing heme active site exhibits strong multiconfigurational electronic character that makes accurate classical simulation exceptionally challenging (Goings et al., 2022). This example is drawn from the resource estimation study of Goings et al. (Goings et al., 2022), which represents one of the most rigorous published analyses of quantum resource requirements for a drug-relevant target.

• Step 1 — Classical Setup (Target Identification and Structure Preparation). Classical MD simulation (AMBER or GROMACS force field) is used to equilibrate the CYP3A4–substrate complex and generate an ensemble of 100–1000 representative conformations. Protein preparation, protonation state assignment, and active space selection are performed classically using standard quantum chemistry software (e.g., ORCA, Gaussian). This step has no quantum hardware requirement. Estimated classical compute: 10–100 GPU-hours per conformation ensemble.
• Step 2 — Active Space Definition (Classical–Quantum Interface). The CYP3A4 active site requires an active space of approximately 50 orbitals (50 electrons in 50 orbitals, denoted CAS(50e,50o)) to correctly describe the iron–oxo reaction intermediates and spin-state manifold (Goings et al., 2022). Classical CASSCF at this active space size is computationally intractable (∼10³⁰ determinants). The active space Hamiltonian is mapped onto qubits using a Jordan–Wigner or Bravyi–Kitaev encoding, requiring approximately 100 logical qubits for a minimal active space of 20–30 orbitals. For the full 50-orbital active space, approximately 100–200 logical qubits are needed.
• Step 3 — NISQ-Era Option: VQE with Error Mitigation. On a current NISQ device (e.g., IBM Eagle, 127 physical qubits), VQE with a UCCSD ansatz can be applied to a reduced active space of 12–16 qubits (12–16 orbitals). Circuit depth for a 12-qubit UCCSD ansatz is approximately 10²–10³ two-qubit gates. With current gate error rates of ∼10⁻³ per two-qubit gate, circuits of this depth accumulate significant noise; zero-noise extrapolation (ZNE) or probabilistic error cancellation (PEC) can partially mitigate errors but introduce a sampling overhead of 10²–10⁴× per expectation value. This approach yields qualitative electronic structure information for model active sites but cannot yet provide the chemical accuracy (<1 kcal/mol) needed for reliable binding affinity prediction at the full CYP3A4 scale.
• Step 4 — FTQC Target: QPE for Chemical Accuracy. Fault-tolerant QPE on the full CAS(50e, 50o) Hamiltonian was originally estimated to require 2,000–10,000 logical qubits and 10⁸–10¹⁰ T-gates (Goings et al., 2022; Cortes et al., 2024). Recent advances in tensor factorisation and active volume compilation (see also pre-print work arXiv:2502.15882 (Google Quantum AI et al., 2025)) suggest T-gate counts for comparable active spaces may be reduced by one to two orders of magnitude, though millions of physical qubits are still required. Critically, this workflow addresses only a single molecular geometry (the compound I iron-oxo intermediate of CYP3A4); a full treatment of drug binding requires conformational ensemble sampling—additional quantum or classical steps not captured here. Full CYP3A4 QPE therefore remains a post-2030 target, though compilation advances may progressively shorten this timeline.
• Step 5 — Classical Post-Processing (Lead Optimisation). Quantum-computed energies for the QM region are passed back to the classical QM/MM framework. The remaining protein environment (MM region, ∼50,000–100,000 atoms) is treated with a molecular mechanics force field. Free energy contributions from the MM region (electrostatics, solvation, conformational entropy) are computed classically via thermodynamic integration or FEP. The final binding affinity is assembled as: ΔG_pea = ΔG_pea (QM) + ΔG_pea (MM) + ΔG_pea (QM/MM coupling). The entire workflow runs on a hybrid classical–quantum infrastructure, with quantum hardware invoked only for the active space energy evaluation in Steps 3–4.

This worked example illustrates that the quantum component of hybrid drug discovery workflows is modular and well-defined: it replaces a specific, computationally intractable subroutine (high-accuracy multireference electronic structure for a transition-metal active site) while the remainder of the pipeline remains classical. The key bottleneck to near-term deployment is not algorithmic but hardware: reducing physical qubit error rates from the current ∼10⁻³ to ∼10⁻⁶ (required for surface-code logical qubits at practical overhead), and increasing qubit counts by three to four orders of magnitude relative to today’s NISQ devices.

9 Challenges in QC for drug discovery

While promising, quantum computing in drug discovery still has considerable hurdles to overcome, both in terms of hardware, software, and algorithms. One convenient classification is one between near-term NISQ-era capabilities and longer-term targets for fault-tolerant quantum computing (FTQC). Variational algorithms like VQE for small molecules, quantum machine learning classifiers and quantum enhanced optimization (with heuristic performance) work with NISQ devices (roughly 50–1000 noisy qubits, 2020–2030 timeframe). FTQC devices (thousands to millions of error-corrected logical qubits, post-2030) will enable Quantum Phase Estimation (QPE) for exact ground-state energies, fault-tolerant simulation of drug-sized molecules (50+ atoms), and Grover-based database search. The two most significant near-term challenges are error mitigation and scalability.

9.1 Scalability and error mitigation

The electronic structure of drug-relevant molecules requires qubits and circuit depths that significantly surpass current capabilities on a quantum computer (Lolur et al., 2023). For example, to encode a 50-orbital active space would require at least 100 logical qubits, which would currently require about 1,000 physical qubits for surface-code error correction, bringing full fault-tolerant simulation of even tinker with really small active space to millions of physical qubits. The parallel bottleneck is circuit depth: the more layers of gates, the more decoherence errors, with molecular simulations for industrially relevant applications in mind, circuit depth is expected to be orders of magnitude larger than what can be done faithfully with current NISQ devices. Finally, error correction is a key issue; quantum systems are highly sensitive to decoherence and errors in the gates reduce the accuracy of simulations. The next step is therefore finding ways to make the devices more accurate so that it would be possible to conduct drug discovery research, as well as creating efficient error correction techniques and establishing powerful quantum algorithms. Further, the conditions surrounding quantum computers are very sensitive, and temperature changes among other factors are likely to result in quantum decoherence and quantum noise that would cause errors, necessitating strict control of conditions for minimum noise and error (Duan et al., 1999). Last but not the least; translating classical drug discovery algorithms to the quantum model is also a significant challenge, and as such, quantum-specific algorithms, which can solve real-life drug discovery problems, need to be designed.

9.2 Hardware and software constraints

Quantum accelerated drug discovery is constrained by hardware development across several fronts at the same time (Li et al., 2021). Two essential limitations exist for qubit connectivity: first, the current processors support only “near neighbour” couplings and second, the overhead of the SWAP needed to simulate the all-to-all coupling of a molecular Hamiltonian introduces significant gate depth penalties. There are a number of ways the noise enters into the hardware: gate infidelity, idle decoherence during the execution of the circuit, and readout error, each of which causes systematic bias of the energy expectation values that will grow non-linearly with circuit depth and is not completely correctable with classical post-processing. Even more complex molecular simulations have an impact on the practicality of the technology, as the required fidelity and stability are difficult to achieve with the existing technology. Furthermore, the scale and complexity of drug discovery simulations utilizing QC are restricted by quantum volume, a global measure that takes into account the amount of qubits, the fidelity of gate interactions, and their connectedness.

Quantum algorithms specifically designed for drug discovery are also required in the software domain. Knowledge of quantum physics and in-depth understanding of the domain-specific problems in drug development are essential for converting classical algorithms to quantum models (Knill et al., 2001). Furthermore, since the lack of user-friendly and effective quantum software packages can be a barrier to the widespread application of QC in this field, it is crucial to develop a robust quantum software ecosystem that includes compilers, programming languages, and libraries that can be used in drug discovery processes.

9.3 Data security implications for pharmaceutical research

Although cryptography is outside the primary scope of this review, the security implications of quantum computing for pharmaceutical data infrastructure are briefly noted. Fault-tolerant quantum computers running Shor’s algorithm will be capable of breaking RSA and elliptic-curve encryption currently protecting clinical trial data, proprietary molecular datasets, and patient health records (Ugwuishiwu et al., 2020). The pharmaceutical industry should therefore begin evaluating post-quantum cryptographic standards (such as those being standardized by NIST) to protect sensitive drug discovery data in anticipation of the quantum threat (Trifiro and Crisafulli, 2022). Quantum Key Distribution (QKD) offers an information-theoretically secure communication channel, though practical deployment at scale faces engineering challenges. This transition is a data governance consideration that will need to be addressed in parallel with the adoption of quantum computing tools for drug discovery itself.

9.4 Hybrid approaches: combining classical and quantum computing

Hybrid quantum-classical integration is a pragmatic near-term strategy that assigns computationally bottlenecking quantum subroutines to quantum hardware while retaining the surrounding classical pipeline (Fadli et al., 2023). Resource efficiency is critical: each quantum circuit execution consumes limited qubit coherence time and gate budget, so hybrid workflows must minimise the quantum circuit depth and the number of parameter update iterations required to converge. As quantum hardware evolves, the quantum component of these hybrid workflows can be systematically increased over time, offering a natural path of growth without needing to replace existing, proven classical software systems. The appeal of hybrid architectures is that they combine the best of both worlds, with quantum processors being used for the portion of calculation where the power of superposition and entanglement of the quantum system holds an advantage—namely electronic structure and quantum sampling; and classical processors on the tasks for which well-developed and reliable software is established, such as optimisation and data management and molecular mechanics. Moreover, this would optimize the use of the quantum resources which are typically limited and would be strategically placed in a more generic conventional workflow. In summary, QC is largely constrained by hardware and software limitations to be used to speed drug discovery (Leelananda and Lindert, 2016). These constraints need to be addressed to realize the transformative promise of QC in this critical discipline of study.

10 Conclusion

This review has traced the theoretical basis and the near and long-term prospects of using quantum computation in every stage of the pharmaceutical development process. There are four main conclusions. Firstly, the most tractable near-term pathway is the modular embedding of NISQ-era variational algorithms (mainly VQE and related hybrids involving a quantum-classical part) within an otherwise classical CADD workflow. They are presently only applicable to small active spaces, but they already allow the observation of electronic structure at transition-metal active sites that is otherwise hard to obtain classically. Second, achieving CCSD(T)-level accuracy for drug-sized systems of 50 or more atoms requires fault-tolerant quantum processors with millions of physical qubits—a target that hardware roadmaps place beyond 2030. Realising this demands parallel advances in qubit fabrication yield, error-correction code efficiency, and compilation algorithms that reduce logical gate overhead. Third, there are no standardised benchmark protocols to demonstrate the capabilities of QCs in the field of drug discovery. There are known and widely-used sets of chemical test cases, accuracy benchmarks and reproducibility criteria for classical computational chemistry studies of binding affinity, and such an agreed set for quantum chemistry on quantum hardware is lacking and needs to be developed as a community priority. The fourth is integrating quantum technologies into regulated pharmaceutical processes involves a data governance strategy, GMP/GLP traceability compliance, and post-quantum cryptographic infrastructures to secure proprietary molecular data. The TRLs assigned to the various quantum chemistry applications in this review are TRL 2–3 for current applications, TRL 4–5 for near-term NISQ applications, and TRL 1–2 for foreseeable FTQC applications. The journey towards this readiness necessitates a concerted and interdisciplinary effort involving the engineering and development of quantum work, theoretical chemistry, pharmaceutical science, and regulatory matters. The most accurate way to think about quantum computation today is not as a disruptive technology that replaces classical CADD, but rather as a technology that will be built on a solid scientific foundation that will eventually be able to remove the accuracy–scalability paradox that has long constrained drug discovery using classical computing.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

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.

Footnotes

Edited by: Ruby Srivastava, Indian Institute of Technology Bombay, India

Reviewed by: Eugene Radchenko, Lomonosov Moscow State University, Russia

Nagaraju Regonda, Malla Reddy University, India