Computational screening of plant-derived compounds against SARS-CoV-2: methodological trends, limitations, and opportunities

1 Introduction

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the virus responsible for the COVID-19 pandemic, triggered an unprecedented global effort to identify novel antiviral candidates. Although the World Health Organization declared in May 2023 that COVID-19 no longer constituted a public health emergency of international concern, the extensive body of computational studies generated during the emergency phase remains valuable for assessing methodological practices, validation standards, and translational limitations in antiviral lead prioritization. SARS-CoV-2 is an enveloped, positive-sense, single-stranded RNA virus. Its pathogenicity and replication cycle depend on a highly coordinated set of viral proteins that are responsible for host cell entry, genome replication, and immune evasion. The viral genome encodes four essential structural proteins, spike (S), membrane (M), envelope (E), and nucleocapsid (N), and 16 nonstructural proteins (nsps). Key enzymes, such as main protease (M^pro) and papain-like protease (PL^pro), are critical for proteolytic cleavage of polyproteins pp1a and pp1ab. Meanwhile, RNA-dependent RNA polymerase (RdRp) drives genome replication, a process supported by nsp15 and nsp16 that facilitate RNA processing and viral evasion of the host immune response. The rapid determination of high-resolution crystallographic structures of these proteins enabled extensive application of structure-based drug design (SBDD) and virtual screening strategies (Rahmattullah et al., 2021; Lamers and Haagmans, 2022; Carabelli et al., 2023).

Among the different chemical sources explored in the search for novel antivirals against SARS-CoV-2, plant-derived secondary metabolites have attracted significant attention due to their structural diversity and long-standing relevance in drug discovery. This, coupled with the high cost and time requirements of experimental screening, has resulted in numerous in silico studies evaluating phytochemical libraries against SARS-CoV-2 proteins, typically employing techniques such as molecular docking and molecular dynamics (MD) simulations, binding free energy calculations, and pharmacokinetic filtering.

However, the rapid proliferation of these in silico studies has resulted in substantial methodological heterogeneity. The current literature varies greatly in terms of docking protocols, MD simulation timescales, treatment of binding free energies, and inconsistent incorporation of pharmacokinetic or toxicity filters. This variability complicates direct cross-study comparisons and raises concerns about the robustness and translational relevance of many reported predictions. These challenges are exacerbated by the intrinsic properties of plant-derived compounds, whose structural and physicochemical properties often fall outside the applicability domain of conventional drug discovery filters, and also by the biological complexity of SARS-CoV-2 targets, which range from well-defined enzymatic pockets, such as M^pro, to highly dynamic and structurally complex protein-protein interaction interfaces such as the Spike-ACE2 system. As a result, the reliability of computational predictions is not only determined by the chosen computational methods, but also by how these are adapted to the specific characteristics of both the chemical space and the biological target (Guerra et al., 2022).

This work provides a critical analysis of recent in silico screening studies focused on plant-derived compounds targeting SARS-CoV-2 proteins. Rather than summarizing individual compounds, we examine methodological trends and implicit assumptions across reported workflows. By discussing how different computational workflows correlate virtual predictions with in vitro validation, we aim to highlight the recurring strengths and limitations that shape the reliability of current computational practices in the ongoing search for effective antiviral therapeutics.

2 Plant-derived compounds: opportunities and biases

The use of plants as treatment is rooted in traditional medicine, a summary of knowledge, skills, and practices based on experiences of indigenous cultures for treating physical and mental illness (Salmerón-Manzano et al., 2020). This ethnopharmacological knowledge has historically been a basis of the pharmaceutical industry, providing a cost-effective foundation for developing modern therapeutics. The clinical integration of complex specialized metabolites such as paclitaxel, etoposide phosphate, topotecan, and homoharringtonine is evidence of the success of this approach (Marchev et al., 2020).

The structural diversity, stereochemical complexity and high degree of functionalization of plant-derived metabolites enable these molecules to form a wide range of non-covalent interactions with biological targets, making them attractive candidates for virtual screening campaigns (Salmerón-Manzano et al., 2020). However, from a computational and drug-discovery standpoint, this complexity introduces significant challenges and inherent biases. The structural features that enable high affinity binding are often associated with high molecular weights, increased polarity and extensive hydrogen-bonding capacity. These characteristics frequently lead to violations of drug-likeness filters, such as Lipinski’s and Veber’s rules. Polyphenolic and highly oxygenated scaffolds, which are common in plant extracts, are often favored by many scoring functions for protein-ligand docking due to their ability to form extensive hydrogen bonds (Guerra et al., 2022). Beyond these scoring artifacts, many in silico workflows prioritize this predicted binding affinity while overlooking other physicochemical properties of the compounds that can complicate their translational potential, such as poor solubility, limited membrane permeability, and extensive metabolic liabilities (Saliu et al., 2021).

3 Computational workflows in SARS-CoV-2 virtual screening

Computational studies of plant-derived compounds against SARS-CoV-2 targets typically follow a similar SBDD pipeline involving molecular docking studies to predict binding modes, MD simulations to evaluate complex stability, and binding free energy calculations to refine the ranking. Finally, pharmacokinetic filters are applied to evaluate drug-likeness. However, substantial differences exist in how these strategies are implemented across studies. Table 1 summarizes the methodological diversity observed in representative workflows, including differences in target selection, simulation protocols, validation strategies and experimental confirmation, and the software or computational platforms reported for each workflow component.

TABLE 1. Methodological landscape and validation status of plant-derived inhibitors against SARS-CoV-2 targets.

3.1 Molecular docking as the initial screening step

Molecular docking is the primary method used to identify favorable binding modes of plant-derived compounds within SARS-CoV-2 targets. The reliability of docking results depends strongly on protocol validation, which is commonly performed through redocking of co-crystallized ligands to ensure that the software can reproduce experimental poses with a root-mean-square deviation (RMSD) below 2.0 Å (Saliu et al., 2021; Khamto et al., 2023). Although tools such as AutoDock Vina and GOLD are widely used due to their efficiency, newer algorithms incorporating machine learning, such as GNINA, have also been adopted (Al-Mahrami et al., 2024).

However, validation strategies remain inconsistent across studies, particularly for targets lacking experimentally validated binding sites. Moreover, structural validation alone does not guarantee the predictive performance of a docking protocol, as high docking scores do not necessarily indicate stable or biologically relevant interactions. In principle, the ability of a scoring function to distinguish true binders should be evaluated using statistical metrics such as the area under the receiver operating characteristic curves (AUROC). Nevertheless, such validation strategies are rarely implemented in studies involving plant-derived compounds against SARS-CoV-2, which implies that the reliability of docking-based rankings remains largely unquantified.

Another critical limitation of molecular docking is that most protocols treat the proteins as rigid structures, neglecting conformational flexibility within the binding site, which can play a critical role in ligand recognition and stabilization. A direct consequence of this approach is that studies relying solely on docking without subsequent dynamic validation may overestimate the potency of compounds that may be unstable in a physiological environment.

3.2 Molecular dynamics simulations and energetic refinement

To further refine docking results by incorporating protein flexibility and solvent effects, MD is often employed. This stage is useful for identifying potential “false positives” that may arise from docking studies, where ligands initially appear favorable but fail to maintain stable interactions under dynamic conditions. For example, some studies employed simulations ranging from 50 to 100 ns to assess the stability of complexes involving liensinine or vernolepin (Bakheit et al., 2023; Khamto et al., 2023). However, from a critical perspective, short simulations, often below 100 ns, may fail to capture slow conformational transitions, potentially leading to premature conclusions about the persistence of the predicted binding mode.

In many workflows, MD simulations are further complemented by binding free energy calculations such as MM-PBSA or MM-GBSA (Molecular Mechanics with Poisson-Boltzmann/Generalized Born Surface Area). These calculations offer post-docking energetic estimates that can support relative compound ranking within a given workflow. Additional approaches, such as Solvation Interaction Energy (SIE), can further support energetic rescoring. Such approaches allowed the exclusion of compounds such as feruloyl tyramine, which despite favorable initial docking, showed less consistent energetic support in subsequent analyses, acting as an additional prioritization filter before experimental testing (Rajchakom et al., 2025).

3.3 Pharmacokinetic and toxicity filtering strategies

Pharmacokinetic and toxicity predictions are commonly incorporated into computational workflows to evaluate the drug-likeness of candidate compounds. This is often done through ADMET (absorption, distribution, metabolism, excretion, and toxicity) predictions. Integrating ADMET predictions is a critical but inconsistently applied step. As our analysis shows, the timing of these filters significantly alters the outcome of drug discovery. Early filtering restricts libraries to compounds with more favorable pharmacokinetic profiles prior to docking, improving computational efficiency(Saliu et al., 2021; Rajchakom et al., 2025). Conversely, late-stage filtering frequently results in the identification of high-affinity compounds with subsequent poor pharmacokinetic profiles (Kashyap et al., 2022; Bakheit et al., 2023).

4 Methodological variability across SARS-CoV-2 targets

This mini-review was designed as a focused methodological review rather than a systematic review. Peer-reviewed studies published from January 2020 to June 2026 were identified through PubMed and Google Scholar using combinations of terms related to SARS-CoV-2, plant-derived compounds, and computational screening methods, including “SARS-CoV-2”, “COVID-19”, “phytochemical”, “plant-derived”, “medicinal plant”, “natural products”, “molecular docking”, “virtual screening”, “in silico”, “ADMET”. PubMed was used to retrieve biomedical-indexed studies, whereas Google Scholar was used to capture relevant peer-reviewed computational studies published in journals not consistently indexed in PubMed. Studies were selected to provide a representative methodological comparison across SARS-CoV-2 targets rather than an exhaustive catalogue of all published screening studies. We included peer-reviewed articles evaluating plant-derived compounds, phytochemicals, or natural product libraries against defined viral or host-associated targets (M^pro, PL^pro, RdRp, Spike, ACE2, TMPRSS2, nsp15, and nsp16), if they contained sufficient methodological detail to support comparison of computational workflows. Preprints, conference proceedings, review articles without original computational analysis, and studies lacking defined molecular targets or transparent computational protocols were excluded.

While most in silico investigations of plant-derived compounds follow the general workflow described in Section 3, its implementation and reliability vary depending on the structural and functional characteristics of the target proteins in SARS-CoV-2. In this section, we examine how the nature of the molecular target influences the interpretation of the screening results, highlighting target-specific challenges and the reliability of the proposed computational hits.

4.1 Spike protein: challenges in targeting protein-protein interfaces

Early in the pandemic, drug discovery efforts primarily focused on the S protein, a glycosylated homotrimer, due to its essential role in viral entry. The S protein undergoes sophisticated structural changes: initially, chain A is exposed for receptor recognition, while chains B and C remain inaccessible. Receptor binding triggers the detachment of chain A, exposing the entire trimer. Host proteases, such as transmembrane serine protease 2 (TMPRSS2), further regulate the infection process by cleaving the S1/S2 junction and activating the S2 subunit for membrane fusion or endocytosis (Singh et al., 2021).

From a methodological perspective, the Spike-ACE2 interface presents a significant challenge for virtual screening. Unlike the deep catalytic pockets of other SARS-CoV-2 enzymes, the Spike-ACE2 contact region is a large, relatively flat protein-protein interaction (PPI) surface, complicating the definition of a suitable binding site. This challenge has been addressed with different levels of rigor. On one hand, several studies have adopted an experimentally guided strategy (Kashyap et al., 2022; Sabzian-Molaei et al., 2022; Khelef et al., 2024), in which residues involved in the RBD-ACE2 (Receptor-Binding Domain - Angiotensin-Converting Enzyme 2) interaction are identified from crystallographic structures prior to virtual screening, thereby restricting the search space to biologically validated regions of the interface. This strategy reduces computational noise by prioritizing validated biological sites. Conversely, other studies have adopted an exploratory strategy aimed at identifying alternative interaction sites beyond the RBD-ACE2 interface. These approaches include the use of databases such as InterPro to predict potential binding domains outside the experimentally characterized contact region (Al-Mahrami et al., 2024), as well as molecular dynamics simulations to identify transient residue interactions that emerge during the viral entry process (Singh et al., 2021).

The reliability of these in silico workflows can be best assessed by their alignment with experimental data. Among the analyzed literature, the study of Urtica dioica agglutinin (UDA), a small plant lectin, stands out because of its high affinity for ACE2, which was predicted via protein-protein docking and supported by MD simulations. This was later corroborated by enzyme-linked immunosorbent assay (ELISA) and MTT cytotoxicity tests in HEK293 cells (Sabzian-Molaei et al., 2022). Similarly, identifying corilagin with docking and MD simulations yielded a potent IC₅₀ of 2.15 ± 0.13 µM, as validated with a BPS Bioscience kit (Al-Mahrami et al., 2024).

Despite these successful examples, the S protein remains a challenging target due to its extreme conformational flexibility and rapid mutation rate, with a limited transferability of static structural models across variants. Consequently, while blocking viral entry is a valuable early intervention strategy, accurately simulating large PPI surfaces is methodologically demanding (Bakheit et al., 2023).

4.2 Host-directed targets: ACE2 and TMPRSS2

Targeting host cell proteins such as ACE2 and TMPRSS2 has been proposed as a strategy to create a barrier against viral entry. However, this approach faces significant biological constraints due to the physiological role of these proteins in human metabolism and the renin-angiotensin-aldosterone system. From a methodological perspective, the challenge lies in identifying compounds that inhibit viral recognition without disrupting the host’s endogenous functions.

In studies targeting ACE2, compounds such as oleanolic acid and betulinic acid showed favorable docking interactions (Kashyap et al., 2022). Nevertheless, these molecules failed to meet drug-likeness criteria, raising concerns regarding toxicity and pharmacokinetic feasibility, illustrating the limitations of affinity-driven prioritization alone.

A critical limitation of standard in silico approaches is further illustrated by studies involving plant extracts and isolated compounds. For example, from the study of AGA isolated from Christia vespertilionis (Ghazali et al., 2025). Although docking studies predicted AGA to be a potent ACE2 inhibitor, in vitro assays in Vero E6 cells revealed that the isolated compound exhibited significantly lower activity than the methanolic extract. This discrepancy highlights a significant limitation of docking-based screening studies, the inability to account for synergistic effects between multiple phytochemicals present in plant extracts.

Similar methodological concerns apply to TMPRSS2. The lack of experimental structures during the early pandemic forced many studies to rely on homology models for virtual screening. Early predictions suggested interactions with residues such as His296 and Ser441 (Kashyap et al., 2022). However, subsequent crystallographic data (PDB ID: 8YOY) revealed that the actual interaction surface involves a different set of residues, such as Tyr414 and Arg470. This discrepancy shows that computational hypotheses derived from homology models should be treated with extreme caution until they are validated by experimental structures or enzymatic assays. An incorrect binding site definition can invalidate the entire virtual screening process.

4.3 RdRp: methodological divergence in covalent and non-covalent strategies

RdRp is considered a high-priority pharmaceutical target in SARS-CoV-2 due to its essential role in viral replication, high conservation across SARS-CoV-2 variants and the lack of close human homologs. However, studies targeting RdRp illustrate how different methodological assumptions can lead to different screening outcomes.

One strategy involved the use of covalent docking combined with MD simulations to identify potential covalent inhibitors from Baccharoidess schimperi, including vernolepin and vernadolol (Bakheit et al., 2023). While these compounds demonstrated high stability during 50 ns simulations, the irreversible nature of covalent bonding carries an inherent risk of tissue accumulation and systemic toxicity. A rigorous covalent docking workflow should therefore go beyond pose stability and explicitly identify the electrophilic warhead and the targeted nucleophilic residue, incorporate reactivity filtering to exclude overly reactive or promiscuous compounds, and include selectivity profiling to evaluate potential off-target covalent modification of host proteins. Without these additional steps, covalent docking results should be interpreted as preliminary hypotheses rather than validated irreversible inhibition mechanisms.

In contrast, other studies have prioritized non-covalent affinity-based screening, identifying compounds such as hesperidin as a potential multi-target candidate against RdRp and M^pro using docking and MD simulations (Mosquera-Yuqui et al., 2022). However, despite favorable computational stability, hesperidin presents known pharmacokinetic limitations, including low oral bioavailability and extensive metabolic degradation. This example highlights a recurring limitation in many computational workflows, where ADMET assessments are incorporated only at later stages of the screening process.

4.4 M^pro as the most tractable SARS-CoV-2 target

The main protease (M^pro or 3CL^pro) is the most extensively studied SARS-CoV-2 target. It plays an essential role in viral polyprotein processing, it has 96% sequence similarity with the original SARS-CoV, and lacks close human homologs, which significantly reduces the risk of off-target toxicity.

The reliability of M^pro screening is best illustrated by workflows that successfully progressed from virtual screening to enzymatic validation. For example, the screening of 326 Thai natural products against the 6LU7 structure, led to the identification of compounds including neferine, liensinine, and isoliensinine (Khamto et al., 2023). Although these compounds initially violated Lipinski’s rule of five due to their large and structurally complex scaffolds, they passed subsequent drug-likeness evaluations and demonstrated strong inhibitory activity in untagged M^pro enzymatic assays, together with low toxicity in Vero cells (CCL-81). Nevertheless, their atypical physicochemical properties still require careful in vivo evaluation to determine whether limitations in oral bioavailability or systemic safety could compromise their translational potential.

Similarly, integrated workflows combining phytochemical characterization, molecular simulations, and enzymatic validation have further demonstrated the tractability of M^pro. One such approach used NMR-guided compound identification together with molecular dynamics simulations and MM-GBSA calculations to prioritize luteolin and 4-O-(4′-O-galloyl-α-L-rhamnopyranosyl)-ellagic acid as potential inhibitors from Saudi medicinal plants. These computational predictions were supported by MBP-tagged 3CL protease assays, in which the ellagic acid derivative exhibited the greatest potency without inducing cellular toxicity, as determined by MTT assays (Alharbi et al., 2024). The agreement between MM-GBSA-based prioritization and enzymatic inhibition in this case suggests that binding free-energy calculations can support post-docking ranking in M^pro workflows, although they remain method-dependent and require experimental validation.

Despite the abundance of high-resolution structures such as 6LU7, 6Y84, and 7TLL, significant methodological inconsistencies persist. While some studies employed a detailed pipeline that included LC-HRMS, Vero E6 infection assays, and MD simulations to identify stable binders such as myricetin 3-O-β-D-galactopyranoside (Maikhunthod et al., 2024), other studies lack this level of biological foundation. For example, relying solely on docking with a structure, without enzymatic validation, makes proposed inhibitors, such as chebulagic acid, remain hypothetical (Singirisetty et al., 2025).

Furthermore, M^pro has also served as a useful framework for evaluating the impact of viral variants on virtual screening performance. Comparative docking analyses between the wild-type and Omicron variants enabled the identification of high-affinity phytochemicals from the LOTUS database (Eltaib and Alzain, 2023). However, subsequent ADMET predictions indicated that several of the top-ranked compounds exhibited poor oral bioavailability despite favorable docking scores. This highlights the importance of incorporating effective pharmacokinetic filtering strategies within computational workflows, particularly when structurally complex phytochemicals are prioritized based primarily on predicted affinity. Even for a comparatively tractable target such as M^pro, favorable binding predictions alone are insufficient if pharmacokinetic limitations ultimately compromise translational feasibility.

4.5 PL^pro, nsp15, and nsp16: target-specific challenges in workflow validation

Papain-like protease (PL^pro), endoribonuclease (nsp15), and 2′-O-methyltransferase (nsp16) are essential but understudied targets in the SARS-CoV-2 proteome. PL^pro is a multifunctional cysteine protease that processes the viral polyprotein and suppresses innate immunity through deubiquitination. In contrast, nsp15 is an endoribonuclease involved in viral RNA processing and immune evasion, whereas nsp16 is a 2′-O-methyltransferase that contributes to immune evasion through viral RNA cap methylation.

The study of PL^pro inhibitors illustrate how workflow design can substantially influence compound prioritization. Conventional strategies based on docking followed by MD simulations identified quercetin 3′-glucoside as a stable binder (Ananto et al., 2024). In contrast, more restrictive workflows incorporating early-stage ADMET filtering followed by energetic refinement using MM-GBSA and Solvation Interaction Energy (SIE) calculations enable the prioritization of molecules with more coherent in silico profiles, and the exclusion of compounds such as feruloyl tyramine, which appeared favorable during docking but showedless consistent energetic support in subsequent analysis (Rajchakom et al., 2025).

Methodological limitations have become even more pronounced in studies targeting nsp15. The widely used crystallographic structure 6VWW lacks a co-crystallized inhibitor, preventing standard redocking validation and increasing uncertainty in binding site definition. Consequently, docking predictions of compounds such as epigallocatechin and amentoflavone should be interpreted cautiously in the absence of biochemical confirmation (Singirisetty et al., 2025).

Among these less explored targets, nsp16 represents one of the few examples where the availability of a co-crystallized ligand enables a more direct pose-validation strategy. The use of the 6YZ1 structure, which contains the co-crystallized inhibitor sinefungin, allowed redocking validation prior to virtual screening. The workflow involved extensive pharmacokinetic filtering, including Lipinski, Ghose, Veber, Egan, and Muegge rules, before docking, narrowing a library of 100 phytochemicals to 59 drug-like molecules, prioritizing andrographolide and quercetin as candidate hits. However, even with a robust workflow, quercetin was flagged for mutagenic potential and blood-brain barrier permeability (Saliu et al., 2021).

5 Emerging methodological patterns across computational screening studies

One of the most consistent patterns across the analyzed studies involving plant-derived compounds against SARS-CoV-2 is the strong reliance on docking scores as the primary criterion for compound prioritization, where small differences in predicted binding affinity are interpreted as evidence of superior inhibitory potential, despite the limited precision of docking scoring functions and the absence of rigorous validations. This affinity-driven prioritization is particularly evident in the recurrent selection of flavonoid and flavonoid-derived scaffolds as top-ranked hits across functionally distinct SARS-CoV-2 targets. For instance, quercetin-derived compounds are repeatedly reported among prioritized hits, including various quercetin glycosides identified in studies targeting M^pro (Hossain et al., 2023; Maikhunthod et al., 2024) and quercetin 3′-glucoside against PL^pro (Ananto et al., 2024). Similarly, hesperidin has been proposed as a multitarget candidate against both M^pro and RdRp (Mosquera-Yuqui et al., 2022). Other representative examples include catechin gallate and amentoflavone-related biflavonoids among M^pro hits (Abdizadeh et al., 2023; Hossain et al., 2023), as well as quercetin prioritized among nsp16 candidates(Saliu et al., 2021). While this recurrence may suggest genuine antiviral potential for some compounds, it may also arise from compound-library composition and docking scoring biases, since highly oxygenated aromatic scaffolds can be favored by scoring functions that reward hydrogen bonding, π-interactions, and extensive protein-ligand contact surfaces. Consequently, the repeated prioritization of these compounds across functionally distinct targets should be interpreted with caution and requires experimental validation before being considered evidence of broad-spectrum antiviral activity.

Another recurring pattern is the imbalance between computational complexity and experimental validation. While many studies progressively increase the sophistication of in silico analysis through MD simulations and free energy calculations, only a limited number proceed toward enzymatic assays or cellular validation. Consequently, increasing computational refinement does not necessarily translate into greater biological confidence or translational relevance.

6 Conclusions and future perspectives

The rapid expansion of computational screening studies during the COVID-19 pandemic demonstrated the potential of in silico approaches for the exploration of plant-derived chemical space. However, the substantial methodological variability observed across studies also highlights the need for more standardized and critically validated workflows.

One important improvement involves the incorporation of validation strategies beyond structural redocking. Whenever possible, docking protocols should be evaluated using benchmark datasets containing active and inactive compounds, allowing assessment through statistical metrics such as the AUROC. These approaches would provide a more quantitative measure of predictive performance and improve confidence in docking-based prioritization.

Another critical consideration is the integration of pharmacokinetic filtering at earlier stages of the workflow. Although late-stage ADMET evaluation allows broader exploration of chemical diversity, it frequently results in the prioritization of compounds with limited translational potential. Early incorporation of pharmacokinetic criteria may improve computational efficiency and reduce the progression of unsuitable candidates into computationally expensive refinement stages. Moreover, because different ADMET platforms rely on distinct training datasets and predictive models, consensus-based approaches may provide more robust pharmacokinetic assessment for structurally complex phytochemicals.

Experimental observations showing higher antiviral activity for crude extracts compared with isolated compounds also expose an important conceptual limitation of current computational workflows, which predominantly evaluate phytochemicals as individual ligands. While this reductionist strategy simplifies virtual screening, it may fail to capture synergistic or cooperative effects arising from complex phytochemical mixtures. How to appropriately represent multi-component plant systems within structure-based screening frameworks remains an open methodological question.

A related limitation is that most reviewed studies focus on isolated protein-ligand interactions and therefore provide limited information on broader molecular pathways. The targets discussed here mainly correspond to major biological processes in SARS-CoV-2 infection, including viral entry, proteolytic processing, RNA replication and processing, and immune evasion. Future studies could integrate target-based screening with network pharmacology or pathway-enrichment analyses to determine whether plant-derived compounds also modulate host inflammatory, innate immune, or oxidative-stress pathways.

Overall, the current literature suggests that the future of antiviral bioprospecting does not lie in the massive generation of docking scores, but rather in the implementation of workflows that prioritize rigorous validation, early-stage pharmacokinetic assessment, and critical interpretation of computational outputs. The recurrent identification of similar phytochemical scaffolds across unrelated SARS-CoV-2 targets further emphasizes the need to distinguish genuine antiviral potential from methodological convergence driven by shared computational biases. Although plant-derived compounds remain a valuable source of structural diversity, their translational potential will only be realized if computational hypotheses are systematically challenged against pharmacokinetic constraints and experimental validation. In this context, virtual screening should be viewed not as a definitive predictor of antiviral activity, but as a hypothesis-generation framework whose value ultimately depends on biological confirmation and methodological rigor.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Generative AI tools were used to support language revision, manuscript organization, and improvement of writing clarity. The authors take full responsibility for the scientific content, interpretation of the literature, critical analysis, and conclusions presented in this work.

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: Vinícius Gonçalves Maltarollo, Federal University of Minas Gerais, Brazil

Reviewed by: Ravi Kumar, Petrichor Biologics foundation, India