Topics and Trends in Most Cited Computational Drug Discovery Methods Papers

Ranked by citations 18 months after publication

Class of 2026 (Papers Published in 2024)

What topics and trends defined most-cited Computational Drug Discovery Methods research in the Class of 2026?

Computational drug discovery is rapidly transitioning from static screening to multi-objective generative AI. De novo molecular design surged fivefold to lead the cohort, supported by graph neural networks and 3D diffusion models. Concurrently, physics-informed molecular dynamics and automated ADMET profiling expanded significantly, while traditional standalone drug-target prediction declined.

See details ↓

Updated July 22, 2026

At a glance

Field
Computational Drug Discovery Methods
Cohort label
Class of 2026 (2024 publications)
Papers analyzed
13,632
Papers ranked
20
Top topics in ranked papers
De novo molecular design, virtual screening, binding affinity prediction, graph neural networks
Publication window
Jan 1, 2024 – Dec 31, 2024
Eligibility
Research articles; reviews excluded
Citation window
18 months post-publication
18m citation range
86–583
Data source
OpenAlex · Retrieved July 2026
License
CC BY 4.0

Rankings

20 papers ranked by 18-month citation count

#1 of 13,632
58318m citations

ProTox 3.0: a webserver for the prediction of toxicity of chemicals

Priyanka Banerjee et al.Nucleic Acids Research202410.1093/nar/gkae303

Priyanka BanerjeeCharité - University Medicine Berlin, Germany

ProTox 3.0Toxicity predictionMolecular similarityMachine learning modelsacute toxicityorgan toxicityclinical toxicitymolecular-initiating eventsTox21 pathwaysadverse outcomestoxicity off-targetsexternal validationChemical structuretoxicity endpointsConfidence scoretoxicity radar plottoxicity network plotWeb serverRisk assessment
Badge
#2 of 13,632
50818m citations

Practical guide to <scp>SHAP</scp> analysis: Explaining supervised machine learning model predictions in drug development

Ana Victoria Ponce-Bobadilla et al.Clinical and Translational Science202410.1111/cts.70056

Sven StodtmannAbbVie Deutschland GmbH & Co. KG, Germany

SHAPexplainable AIModel interpretabilityfeature attributionsupervised machine learningblack-box modelsdrug development applicationsregression modelsclassification modelsbinary endpointstime-series modelsmodel transparencyclinical decision supportfeature importance analysis
Badge
#3 of 13,632
44618m citations

ADMETlab 3.0: an updated comprehensive online ADMET prediction platform enhanced with broader coverage, improved performance, API functionality and decision support

Li Fu et al.Nucleic Acids Research202410.1093/nar/gkae236

Tingjun HouCentral South University, China

ADMETPhysicochemical propertiesmedicinal chemistryD-MPNNDirected message-passing neural networksMolecular descriptorsAPI functionalityuncertainty estimatesDrug discoveryabsorptiondistributionmetabolismexcretionToxicity predictionWeb servercandidate compound selectionPharmacokineticsIn silico prediction
Badge
#4 of 13,632
26518m citations

SwissDock 2024: major enhancements for small-molecule docking with Attracting Cavities and AutoDock Vina

Marine Bugnon et al.Nucleic Acids Research202410.1093/nar/gkae300

Olivier MichielinSIB Swiss Institute of Bioinformatics, Switzerland

SwissDockMolecular dockingAttracting CavitiesAutoDock VinaEADock DSSSwissDrugDesignligand preparationtarget preparationSMILESPDB filesPDB IDcovalent ligand dockingDrug-target interactiondocking accuracyweb-based docking servicecommand-line interfaceVirtual screening
Badge
#5 of 13,632
18018m citations

Leveraging large language models for predictive chemistry

Kevin Maik Jablonka et al.Nature Machine Intelligence202410.1038/s42256-023-00788-1

Berend SmitFriedrich Schiller University Jena, Germany

GPT-3large language modelsFine-tuningnatural language processing for chemistrymolecular property predictionmaterials property predictionchemical reaction yield predictioninverse molecular designLow-data regimeFoundation modelTextual molecular representationtransfer learning in chemistryquestion-answering formatsmall chemical datasetsbenchmarking against specialized modelsnatural language queries for chemistry
Badge
#6 of 13,632
16418m citations

Structure-based drug design with equivariant diffusion models

Arne Schneuing, Charles Harris, Yuanqi Du et al.Nature Computational Science202410.1038/s43588-024-00737-x

Arne Schneuing, Bruno E. CorreiaÉcole Polytechnique Fédérale de Lausanne, Switzerland

Structure-based drug designEquivariant diffusion models3D conditional generationDiffSBDDDe novo molecular designProtein pocket conditioningproperty optimizationnegative designmolecular inpaintingBinding affinity predictionProtein-ligand complexespretrained diffusion modelconstraint-based generation
Badge
#7 of 13,632
15818m citations

Open Targets Platform: facilitating therapeutic hypotheses building in drug discovery

Annalisa Buniello et al.Nucleic Acids Research202410.1093/nar/gkae1128

Annalisa BunielloOpen Targets, United Kingdom

Open Targets Platformdrug target identificationtarget prioritisationtarget-disease associationstherapeutic hypothesesclinical precedenceTractabilitysafety attributesdirection of effect assessmentgenetic variationTarget modulationRisk assessmentknowledge baseDrug discoverytarget annotationevidence sourcesmechanism of modulation
Badge
#8 of 13,632
15518m citations

An artificial intelligence accelerated virtual screening platform for drug discovery

Guangfeng Zhou, Domnita-Valeria Rusnac et al.Nature Communications202410.1038/s41467-024-52061-7

Ning Zheng, Frank DiMaioUniversity of Washington, United States

RosettaVSVirtual screeningMolecular dockingBinding affinity predictionReceptor flexibilityCompound librariesKLHDC2ubiquitin ligaseNav1.7Artificial intelligence in drug discoverymicromolar binding affinitiesX-ray crystallographyHit identificationlead discoveryOpen-source framework
Badge
#9 of 13,632
15118m citations

VFDB 2025: an integrated resource for exploring anti-virulence compounds

Siyu Zhou et al.Nucleic Acids Research202410.1093/nar/gkae968

Siyu ZhouChinese Academy of Medical Sciences & Peking Union Medical College, China

Anti-virulence therapyVirulence factorMultidrug resistanceDrug repurposingtarget selectionLiterature miningChemical structuremolecular targetsMechanism of actionbacterial pathogenesisantibiotic resistanceselective pressureantibacterial therapies
Badge
#10 of 13,632
14618m citations

ADMET-AI: a machine learning ADMET platform for evaluation of large-scale chemical libraries

Kyle Swanson et al.Bioinformatics202410.1093/bioinformatics/btae416

Kyle Swanson, Rabindra V. Shivnaraine, James ZouStanford University, United States

ADMETmachine learningCompound librarieshigh-throughput dockinggenerative AIDrug-likenessabsorptiondistributionmetabolismexcretionToxicity predictionTDC ADMET Leaderboardcombinatorial chemical spacesbatch predictionPython packageWeb serverSmall molecule drug discovery
Badge
#11 of 13,632
14618m citations

Artificial intelligence alphafold model for molecular biology and drug discovery: a machine-learning-driven informatics investigation

Song-Bin Guo, Yuan Meng, Liteng Lin et al.Molecular Cancer202410.1186/s12943-024-02140-6

Song‐Bin Guo, Weijuan HuangSun Yat-Sen University Cancer Center, China

AlphaFoldProtein structure predictionartificial intelligenceDrug discoveryMolecular dynamics simulationscientometric analysismachine-learning-driven informaticsunsupervised clustering algorithmWalktrap algorithmhotspot burst analysisSARS-CoV-2COVID-19vaccine designhomology modelingVirtual screeningmembrane proteinstructural biologytime series trackingglobal impact assessment
Badge
#12 of 13,632
14418m citations

A foundation model for clinician-centered drug repurposing

Kexin Huang, Payal Chandak et al.Nature Medicine202410.1038/s41591-024-03233-x

Marinka ŽitnikHarvard Medical School, United States

TxGNNDrug repurposingZero-shot learningFoundation modelBiomedical knowledge graphsGraph neural networkmetric learningdrug indicationsdrug contraindicationsinterpretable rationalesmulti-hop reasoningExplainer moduleoff-label prescriptionsdiseases with limited treatment optionsbenchmark comparison
Badge
#13 of 13,632
12518m citations

Deep-PK: deep learning for small molecule pharmacokinetic and toxicity prediction

Yoochan Myung et al.Nucleic Acids Research202410.1093/nar/gkae254

Alex G. C. de Sá, David B. AscherThe University of Queensland, Australia

Deep-PKPharmacokineticsToxicity predictionADMETGraph neural networkGraph-based featuresMolecular optimization73 endpointsSmall molecule drug discoveryHigh-throughput screeningabsorptiondistributionmetabolismexcretionModel interpretabilityArtificial intelligence in drug discovery
Badge
#14 of 13,632
11218m citations

AlphaFold predictions of fold-switched conformations are driven by structure memorization

Devlina Chakravarty et al.Nature Communications202410.1038/s41467-024-51801-z

Lauren L. PorterNational Institutes of Health, United States

AlphaFoldfold-switching proteinsstructure memorizationProtein energeticsconformational diversityAF2AF3Confidence scoretraining-set structurescoevolutionary restraintssecondary structure transitionstertiary structure transitionsProtein structure predictionConformational samplingPhysics-based methodsmodel discriminationhigh energy conformationslow energy conformations
Badge
#15 of 13,632
9918m citations

Reinvent 4: Modern AI–driven generative molecule design

Hannes H Loeffler et al.Journal of Cheminformatics202410.1186/s13321-024-00812-5

Hannes H. LoefflerAstraZeneca, Sweden

REINVENT 4generative AIComputational drug designrecurrent neural networksTransformertransfer learningreinforcement learningcurriculum learningDe novo molecular designR-group replacementLibrary designLinker designScaffold hoppingMolecular optimizationTOML configurationJSON configurationDrug discoveryOpen-source frameworkApache 2.0 license
Badge
#16 of 13,632
9318m citations

Machine learning-aided generative molecular design

Yuanqi Du et al.Nature Machine Intelligence202410.1038/s42256-024-00843-5

Philippe Schwaller, Tom L. BlundellCornell University, United States

machine learningDe novo molecular designDeep learning modelsChemical space explorationSMILESmolecular property predictionreinforcement learninggenerative adversarial networksvariational autoencodersrecurrent neural networksMolecular optimizationDrug-likenessSynthetic accessibilitytransfer learning
Badge
#17 of 13,632
9118m citations

Equivariant 3D-conditional diffusion model for molecular linker design

Ilia Igashov et al.Nature Machine Intelligence202410.1038/s42256-024-00815-9

Bruno E. CorreiaÉcole Polytechnique Fédérale de Lausanne, Switzerland

DiffLinkerEquivariant diffusion modelsthree-dimensional conditional diffusionLinker designfragment-based drug discoverydisconnected molecular fragmentsarbitrary number of fragmentsautomatic atom placementattachment point predictionProtein pocket conditioningSynthetic accessibilityChemical diversityDe novo molecular designequivariant neural networksStructure-based drug design
Badge
#18 of 13,632
8818m citations

DynamicBind: predicting ligand-specific protein-ligand complex structure with a deep equivariant generative model

Wei Lu, Jixian Zhang, Shuangjia Zheng et al.Nature Communications202410.1038/s41467-024-45461-2

Wei Lu, Jixian Zhang, Shuangjia ZhengGalixir Technologies, China

DynamicBindligand-specific protein conformationsEquivariant diffusion modelsprotein-ligand complex structure predictionenergy landscape constructionunbound protein structuresMolecular dockingVirtual screeningcryptic pocketsProtein conformational changesdeep equivariant generative modelequilibrium state transitionsholo-structure-free predictionUndruggable targetsprotein dynamics modelingLigand-induced conformational changes
Badge
#20 of 13,632
8618m citations

AlphaFold2 structures guide prospective ligand discovery

Jiankun Lyu, Nicholas Kapolka, Ryan Gumpper, Assaf Alon, Liang Wang et al.Science202410.1126/science.adn6354

Jiankun Lyu, Nicholas J. Kapolka, Ryan H. Gumpper, Assaf Alon, Liang Wang, Georgios Skiniotis, Andrew C. Kruse, Brian K. Shoichet, Bryan L. RothUniversity of California, San Francisco, United States

AlphaFold2Structure-based drug designprospective ligand discoveryVirtual screeningσ2 receptor5-HT2A receptorMolecular dockingorthosteric sitecryo-electron microscopyhit rateBinding affinity predictionReceptor flexibilityunrefined AlphaFold2 modelsexperimental structure comparisonresidue accommodationGPCR structure prediction
Badge

Dominant research themes and year-over-year shifts in Computational Drug Discovery Methods

What Topics Define the Class of 2026?

The Class of 2026 in computational drug discovery reflects a major pivot toward generative molecular design and AI-driven predictive screening. De novo molecular design emerges as the single most dominant theme, present in 20% of high-impact papers (10 out of 50). Classical structure-based methodologies—including molecular docking, virtual screening, and binding affinity prediction—remain fundamental pillars, each featured in 18% of ranked studies. However, these traditional approaches are increasingly augmented by deep learning frameworks such as graph neural networks (14%) and 3D equivariant diffusion models (8%). A key hallmark of this cohort is the holistic integration of property optimization early in the generative pipeline. Rather than treating candidate selection as a post-hoc filtering step, top-ranked studies embed ADMET predictions—spanning absorption, distribution, metabolism, and toxicity—directly into molecular generation workflows. Additionally, machine learning interatomic potentials (8%) and high-throughput molecular dynamics simulations (10%) are expanding the accuracy of conformational sampling and free energy landscapes. Together, these trends demonstrate that computational drug discovery has evolved from static virtual screening toward dynamic, multi-objective generative frameworks capable of designing synthetically accessible, drug-like candidates.

Leading research themes
Leading research themes

How Did Topics Shift from the Class of 2025 to the Class of 2026?

Comparing the Class of 2025 to the Class of 2026 reveals a decisive structural shift from single-target hit identification toward end-to-end generative design. The most dramatic surge occurred in de novo molecular design, which expanded fivefold from 4% of papers in 2023 to 20% in 2024. Next-generation deep learning architectures experienced rapid adoption: graph neural networks grew 2.33-fold (6% to 14%), while 3D equivariant diffusion models and machine learning interatomic potentials emerged as new frontline methodologies, appearing in 8% of ranked publications. Predictive biophysics and safety profiling also recorded significant gains. Molecular dynamics simulations grew 2.5-fold (4% to 10%), while comprehensive ADMET evaluations and conformational sampling quadrupled (2% to 8%). In contrast, legacy methodologies such as standalone drug-target interaction modeling dropped from 26% to 14%, and basic hit identification declined from 10% to 4%. This shift indicates that the field is moving away from isolated binding predictions toward unified AI architectures that simultaneously optimize target binding affinity, synthetic accessibility, and pharmacokinetic profiles in a single computational workflow.

How topics shifted year over year
How topics shifted year over year
Methodology

PRI identifies high-impact research using a transparent, topic-agnostic framework applied consistently across scientific domains. Bibliographic records are drawn from OpenAlex, including publication dates, citation relationships, and document types.

This ranking covers the Class of 2026 cohort: journal articles published in 2024. Reviews and other non-article document types are excluded to ensure comparability.

Research impact is quantified with an 18-month post-publication citation window—the number of citing works published within 18 months of each paper's publication date. This metric captures early impact while controlling for publication age.

An LLM-based relevance classifier then reviews each candidate's title and abstract to confirm substantive alignment with the target domain. Only papers classified as relevant appear in the final ranking.

Zheng Su, Tinsley Li, Thematic Shifts in Early-High-Impact Cancer Genomics and Diagnostics Research: A Bibliometric and Semantic Analysis. bioRxiv 2026.07.04.736459; doi: https://doi.org/10.64898/2026.07.04.736459

Full methodologyGitHubbioRxiv preprint

Cite this ranking

Pepkio Research Index (PRI). Topics and Trends in Most Cited Computational Drug Discovery Methods Papers, Class of 2026. https://pri.pepkio.com/top-papers/computational-drug-discovery-methods/2026. Accessed 2026-07-22.

Zheng Su, Tinsley Li, Thematic Shifts in Early-High-Impact Cancer Genomics and Diagnostics Research: A Bibliometric and Semantic Analysis. bioRxiv 2026.07.04.736459; doi: https://doi.org/10.64898/2026.07.04.736459

Source data

The full ranking corpus and analysis files are openly available on an external repository. Please cite the dataset below when reusing this data.

View source dataset →

Pepkio Research Index (2026). Computational Drug Discovery Methods — Top Papers (Class of 2026) [Data set]. Figshare.