Comparison of structure- and ligand-based scoring functions for deep generative models: a GPCR case study
Open accessPublished version of a preprint
Abstract
Deep generative models have shown the ability to devise both valid and novel chemistry, which could significantly accelerate the identification of bioactive compounds. Many current models, however, use molecular descriptors or ligand-based predictive methods to guide molecule generation towards a desirable property space. This restricts their application to relatively data-rich targets, neglecting those where little data is available to sufficiently train a predictor. Moreover, ligand-based approaches often bias molecule generation towards previously established chemical space, thereby limiting their ability to identify truly novel chemotypes. In this work, we assess the ability of using molecular docking via Glide—a structure-based approach—as a scoring function to guide the deep generative model REINVENT and compare model performance and behaviour to a ligand-based scoring function. Additionally, we modify the previously published MOSES benchmarking dataset to remove any induced bias towards non-protonatable groups. We also propose a new metric to measure dataset diversity, which is less confounded by the distribution of heavy atom count than the commonly used internal diversity metric. With respect to the main findings, we found that when optimizing the docking score against DRD2, the model improves predicted ligand affinity beyond that of known DRD2 active molecules. In addition, generated molecules occupy complementary chemical and physicochemical space compared to the ligand-based approach, and novel physicochemical space compared to known DRD2 active molecules. Furthermore, the structure-based approach learns to generate molecules that satisfy crucial residue interactions, which is information only available when taking protein structure into account. Overall, this work demonstrates the advantage of using molecular docking to guide de novo molecule generation over ligand-based predictors with respect to predicted affinity, novelty, and the ability to identify key interactions between ligand and protein target. Practically, this approach has applications in early hit generation campaigns to enrich a virtual library towards a particular target, and also in novelty-focused projects, where de novo molecule generation either has no prior ligand knowledge available or should not be biased by it.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
These algorithms can be interfaced with docking calculations to design small molecules with high docking scores.
“Of note, these algorithms can be interfaced with docking calculations in order to design small molecules with high docking scores.155,156”
From Advancing Targeted Protein Degradation via Multiomics Profiling and Artificial Intelligence (2023) · cites this paper as [156]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 1× · see citing works
“This filtering resulted in 2,453,916 unique, non-isomeric (stereochemistry removed) SMILES that was subsequently used to train the Prior network for a total of 5 epochs with a batch size of 128 using the Adam optimizer [70] with a learning rate of 0.”
MethodBerman, H. M. (2000) · Nucleic Acids Research · doi:10.1093/nar/28.1.235 · Cited by 40,196 · see citing works
Trott, O., & Olson, A. J. (2010) · Journal of Computational Chemistry · doi:10.1002/jcc.21334 · Cited by 38,412 · cited 2× · see citing works
“Lastly, Boitreaud et al. [55] recently used a novel sampling approach combined with a graph to SELFIES [56] variational autoencoder, where the authors demonstrated the ability to optimize the Vina [34] docking score against Dopamine Receptor D3, while maintaining chemical diversity.”
MethodFriesner, R. A., Banks, J. L., Murphy, R. B., Halgren, T. A., Klicic, J. J., Mainz, D. T., Repasky, M. P., Knoll, E. H., Shelley, M., Perry, J. K., et al. (2004) · Journal of Medicinal Chemistry · doi:10.1021/jm0306430 · Cited by 10,057 · cited 2× · see citing works
“We used this approach to optimize de novo molecules to minimize the docking score returned by Glide [32].”
MethodWeininger, D. (1988) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci00057a005 · Cited by 7,436 · cited 14× · see citing works
“Overlap analysis of molecules generated de novo compared to known active and inactive molecules To assess recovery of known active molecules we identified whether any of the canonical SMILES produced by either Agent matches those of known DRD2 active molecules.”
BackgroundInfluentialUMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
McInnes, L., Healy, J., & Melville, J. (2018) · arXiv · doi:10.48550/arxiv.1802.03426 · Cited by 7,101 · cited 5× · see citing works
“For property space, the MolLogP, MolWt, HeavyAtomCount, NumHAcceptors, NumHDonors, NumHeteroatoms, NumRotatableBonds, NumAromaticRings, NumAliphaticRings, RingCount, TPSA, FractionCSP3, QED [94] and SAscore [95] were calculated using RDKit and scaled before input to UMAP using default parameters.”
MethodInfluentialDevelopment and validation of a genetic algorithm for flexible docking 1 1Edited by F. E. Cohen
Jones, G., Willett, P., Glen, R. C., Leach, A. R., & Taylor, R. (1997) · Journal of Molecular Biology · doi:10.1006/jmbi.1996.0897 · Cited by 6,763 · see citing works
Sastry, G. M., Adzhigirey, M., Day, T., Annabhimoju, R., & Sherman, W. (2013) · Journal of Computer-Aided Molecular Design · doi:10.1007/s10822-013-9644-8 · Cited by 6,130 · see citing works
Docking and scoring in virtual screening for drug discovery: methods and applications
Kitchen, D. B., Decornez, H. Y., Furr, J. R., & Bajorath, J. (2004) · Nature Reviews Drug Discovery · doi:10.1038/nrd1549 · Cited by 4,245 · see citing works
Baell, J. B., & Holloway, G. A. (2010) · Journal of Medicinal Chemistry · doi:10.1021/jm901137j · Cited by 3,963 · cited 1× · see citing works
“5; no charged atoms; no atoms besides C, N, S, O, F, Cl, Br, H; no cycles larger than 8 members; custom medicinal chemistry filters [61, 62]; and finally PAINS filters [63] were applied.”
BackgroundPubChem 2019 update: improved access to chemical data
Kim, S., Chen, J., Cheng, T., Gindulyte, A., He, J., He, S., Li, Q., Shoemaker, B. A., Thiessen, P. A., Yu, B., et al. (2019) · Nucleic Acids Research · doi:10.1093/nar/gky1033 · Cited by 3,249 · cited 1× · see citing works
“ExCAPE-DB is a curation of ChEMBL20 [68] and PubChem [69] data that classifies molecules with a measured dose–response value equal to or lower than 10 μM as active, and with higher than 10 μM (or those which were labelled inactive in the original sources) as inactive.”
BackgroundZINC 15 – Ligand Discovery for Everyone
Sterling, T., & Irwin, J. J. (2015) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.5b00559 · Cited by 3,224 · cited 9× · see citing works
“This is likely partially due to the restriction in molecular weight imposed by the ZINC subset used to train the Prior, which selects molecules with a molecular weight between 250 and 350 Daltons, subsequently biasing de novo molecule generation to a similar molecular weight range.”
MethodInfluentialAutomatic Chemical Design Using a Data-Driven Continuous Representation of Molecules
Gómez-Bombarelli, R., Wei, J. N., Duvenaud, D., Hernández-Lobato, J. M., Sánchez-Lengeling, B., Sheberla, D., Aguilera-Iparraguirre, J., Hirzel, T. D., Adams, R. P., & Aspuru-Guzik, A. (2018) · ACS Central Science · doi:10.1021/acscentsci.7b00572 · Cited by 3,106 · cited 1× · see citing works
“[52] evidenced that CVAE [10] and GVAE [53] were unable to generate molecules with optimized Smina [43] docking scores due to the inaccurate prediction of said docking score, which is used to guide de novo sampling in the respective methods.”
MethodTrends in GPCR drug discovery: new agents, targets and indications
Hauser, A. S., Attwood, M. M., Rask‐Andersen, M., Schiöth, H. B., & Gloriam, D. E. I. (2017) · Nature Reviews Drug Discovery · doi:10.1038/nrd.2017.178 · Cited by 2,609 · cited 1× · see citing works
“More generally, G protein-coupled receptors (GPCRs)—including DRD2—are the most commonly targeted protein class accounting for approximately 34% of all FDA approved drugs [46].”
BackgroundThe Properties of Known Drugs. 1. Molecular Frameworks
Bemis, G. W., & Murcko, M. A. (1996) · Journal of Medicinal Chemistry · doi:10.1021/jm9602928 · Cited by 2,501 · see citing works
A Deep Learning Approach to Antibiotic Discovery
Stokes, J. M., Yang, K., Swanson, K., Jin, W., Cubillos-Ruiz, A., Donghia, N. M., MacNair, C. R., French, S., Carfrae, L. A., Bloom-Ackermann, Z., et al. (2020) · Cell · doi:10.1016/j.cell.2020.01.021 · Cited by 2,463 · see citing works
Directory of useful decoys, enhanced (DUD-E): better ligands and decoys for better benchmarking.
Mysinger, M. M., Carchia, M., Irwin, J. J., & Shoichet, B. K. (2012) · Journal of Medicinal Chemistry · doi:10.1021/jm300687e · Cited by 2,412 · see citing works
ChEMBL: towards direct deposition of bioassay data
Mendez, D., Gaulton, A., Bento, A. P., Chambers, J., De Veij, M., Félix, E., Magariños, M. P., Mosquera, J. F., Mutowo, P., Nowotka, M., et al. (2019) · Nucleic Acids Research · doi:10.1093/nar/gky1075 · Cited by 2,372 · see citing works
Quantifying the chemical beauty of drugs
Bickerton, G. R., Paolini, G. V., Besnard, J., Mureşan, S., & Hopkins, A. L. (2012) · Nature Chemistry · doi:10.1038/nchem.1243 · Cited by 2,301 · cited 4× · see citing works
“Abbreviations ADAM: Adaptative moment estimation; DRD2: Dopamine receptor D2; FEP: Free energy perturbation; FCD: Fréchet chemnet distance; GPCR: G protein-coupled receptor; QED: Quantitative estimate of drug-likeness; QSAR: Quantitative structure–activity relationship; SAscore: Synthetic accessibility score; SELFIES:…”
MethodResultInfluentialShelley, J. C., Cholleti, A., Frye, L. L., Greenwood, J. R., Timlin, M. R., & Uchimaya, M. (2007) · Journal of Computer-Aided Molecular Design · doi:10.1007/s10822-007-9133-z · Cited by 2,091 · see citing works
Ertl, P., & Schuffenhauer, A. (2009) · Journal of Cheminformatics · doi:10.1186/1758-2946-1-8 · Cited by 1,939 · cited 4× · see citing works
“Abbreviations ADAM: Adaptative moment estimation; DRD2: Dopamine receptor D2; FEP: Free energy perturbation; FCD: Fréchet chemnet distance; GPCR: G protein-coupled receptor; QED: Quantitative estimate of drug-likeness; QSAR: Quantitative structure–activity relationship; SAscore: Synthetic accessibility score; SELFIES:…”
MethodInfluentialSøndergaard, C. R., Olsson, M. H. H., Rostkowski, M., & Jensen, J. H. (2011) · Journal of Chemical Theory and Computation · doi:10.1021/ct200133y · Cited by 1,869 · see citing works
Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules.
Gómez‐Bombarelli, R., Wei, J. N., Duvenaud, D., Hernández-Lobato, J. M., Sánchez-Lengeling, B., Sheberla, D., Aguilera‐Iparraguirre, J., Hirzel, T. D., Adams, R. P., & Aspuru‐Guzik, A. (2018) · Cambridge University Engineering Department Publications Database · doi:10.17863/cam.34134 · Cited by 1,813 · see citing works
Junction Tree Variational Autoencoder for Molecular Graph GenerationPreprint
Jin, W.-G., Barzilay, R., & Jaakkola, T. (2018) · International Conference on Machine Learning · doi:10.48550/arxiv.1802.04364 · Cited by 1,783 · see citing works
hERG potassium channels and cardiac arrhythmia
Sanguinetti, M. C., & Tristani‐Firouzi, M. (2006) · Nature · doi:10.1038/nature04710 · Cited by 1,582 · cited 1× · see citing works
“Furthermore, internal diversity measures molecules active against hERG—a promiscuous target related to cardiotoxicity [87]—as diverse as all molecules reported active against any kinases, any family A GPCR and more diverse than a virtual library designed for diversity.”
BackgroundDeep learning enables rapid identification of potent DDR1 kinase inhibitors
Zhavoronkov, A., Ivanenkov, Y. A., Aliper, A., Veselov, M. S., Aladinskiy, V. A., Aladinskaya, A. V., Terentiev, V. A., Polykovskiy, D. A., Kuznetsov, M. D., Asadulaev, A., et al. (2019) · Nature Biotechnology · doi:10.1038/s41587-019-0224-x · Cited by 1,573 · cited 1× · see citing works
“biased training or fine-tuning) [2, 6, 19], or using machine learning models trained on known bioactive molecules to predict de novo molecule bioactivity (e.”
MethodGenerating Focused Molecule Libraries for Drug Discovery with Recurrent Neural Networks
Segler, M. H. S., Kogej, T., Tyrchan, C., & Waller, M. P. (2018) · ACS Central Science · doi:10.1021/acscentsci.7b00512 · Cited by 1,537 · cited 2× · see citing works
“functions In the REINVENT study [6] the authors evaluated the performance of the SVM model on an undisclosed held-out test set, resulting in an accuracy of 98%, precision of 97% and recall of 82%.”
MethodEnumeration of 166 Billion Organic Small Molecules in the Chemical Universe Database GDB-17
Ruddigkeit, L., van Deursen, R., Blum, L. C., & Reymond, J.-L. (2012) · Journal of Chemical Information and Modeling · doi:10.1021/ci300415d · Cited by 1,505 · cited 1× · see citing works
“2): enumerated virtual libraries of stable molecules up to 17 and 13 heavy atoms (GDB17 [83], GDB13 [84]), characterised molecules with varying bioactivities (ChEMBL28 [82]), a synthetically accessible diversity orientated virtual library (Enamine diverse [85]), synthetically accessible targeted virtual libraries (Ena…”
BackgroundDeep reinforcement learning for de novo drug design
Popova, M., Isayev, O., & Tropsha, A. (2018) · Science Advances · doi:10.1126/sciadv.aap7885 · Cited by 1,379 · cited 1× · see citing works
“using reinforcement learning [7, 8, 20, 21], Bayesian optimization [22] or particle swarm optimization [23].”
MethodMolecular de-novo design through deep reinforcement learning
Olivecrona, M., Blaschke, T., Engkvist, O., & Chen, H. (2017) · Journal of Cheminformatics · doi:10.1186/s13321-017-0235-x · Cited by 1,358 · cited 10× · see citing works
“using reinforcement learning [7, 8, 20, 21], Bayesian optimization [22] or particle swarm optimization [23].”
BackgroundMethodInfluentialOPLS3e: Extending Force Field Coverage for Drug-Like Small Molecules
Roos, K., Wu, C., Damm, W., Reboul, M., Stevenson, J., Lu, C. T., Dahlgren, M. K., Mondal, S., Chen, W., Wang, L., et al. (2019) · Journal of Chemical Theory and Computation · doi:10.1021/acs.jctc.8b01026 · Cited by 1,243 · cited 1× · see citing works
“ligand, cofactors), at pH 7 ± 2 using Epik [72], optimized hydrogen bond assignment at pH 7 using PROPKA [73] and minimized the structure using the OPLS3e force field [74].”
MethodLessons Learned in Empirical Scoring with smina from the CSAR 2011 Benchmarking Exercise
Koes, D. R., Baumgartner, M. P., & Camacho, C. J. (2013) · Journal of Chemical Information and Modeling · doi:10.1021/ci300604z · Cited by 1,150 · cited 4× · see citing works
“Preliminary analysis (data not shown) suggests that this is the case when using Smina as opposed to Glide.”
BackgroundMethodInfluentialSpatial validation reveals poor predictive performance of large-scale ecological mapping models
Ploton, P., Mortier, F., Réjou‐Méchain, M., Barbier, N., Picard, N., Rossi, V., Dormann, C. F., Cornu, G., Viennois, G., Bayol, N., et al. (2020) · Nature Communications · doi:10.1038/s41467-020-18321-y · Cited by 1,000 · see citing works
Self-referencing embedded strings (SELFIES): A 100% robust molecular string representation
Krenn, M., Häse, F., Nigam, A., Friederich, P., & Aspuru‐Guzik, A. (2020) · Machine Learning Science and Technology · doi:10.1088/2632-2153/aba947 · Cited by 917 · cited 3× · see citing works
“Abbreviations ADAM: Adaptative moment estimation; DRD2: Dopamine receptor D2; FEP: Free energy perturbation; FCD: Fréchet chemnet distance; GPCR: G protein-coupled receptor; QED: Quantitative estimate of drug-likeness; QSAR: Quantitative structure–activity relationship; SAscore: Synthetic accessibility score; SELFIES:…”
MethodGuacaMol: Benchmarking Models for de Novo Molecular Design
Brown, N., Fiscato, M., Segler, M. H. S., & Vaucher, A. C. (2019) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.8b00839 · Cited by 895 · cited 3× · see citing works
“TP + FP + TN + TN Model performance and diversity metrics Several metrics were used to assess generative model performance, as used in GuacaMol [78] and MOSES [4] (see Additional file 1).”
Method970 Million Druglike Small Molecules for Virtual Screening in the Chemical Universe Database GDB-13
Blum, L. C., & Reymond, J. (2009) · Journal of the American Chemical Society · doi:10.1021/ja902302h · Cited by 874 · see citing works
Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models
Polykovskiy, D., Zhebrak, A., Sanchez-Lengeling, B., Golovanov, S., Tatanov, O., Belyaev, S., Kurbanov, R., Artamonov, A., Aladinskiy, V., Veselov, M., et al. (2020) · Frontiers in Pharmacology · doi:10.3389/fphar.2020.565644 · Cited by 849 · cited 11× · see citing works
“We (1) first removed known DRD2 actives (according to the ExCAPE-DB [58]) from the MOSES curated [4] ZINC [59] database of small drug-like molecules for use as training data.”
BackgroundMethodInfluentialComparative Assessment of Scoring Functions: The CASF-2016 Update
Su, M., Yang, Q., Du, Y., Feng, G., Liu, Z., Li, Y., & Wang, R. (2018) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.8b00545 · Cited by 838 · cited 1× · see citing works
“Although the resulting free energy score is notoriously inaccurate [35, 36] and the performance of these scoring functions can be highly target-dependent [37], molecular docking consistently results in the early enrichment of known active molecules in virtual libraries compared to random [35] and is a generally-applie…”
BackgroundBetter Informed Distance Geometry: Using What We Know To Improve Conformation Generation
Riniker, S., & Landrum, G. (2015) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.5b00654 · Cited by 765 · cited 1× · see citing works
“Lastly, the NPR1 and NPR2 were calculated using RDKit after first generating 3D conformations using the ETKDG method [93].”
MethodDeep learning for molecular design—a review of the state of the art
Elton, D. C., Boukouvalas, Z., Fuge, M. D., & Chung, P. W. (2019) · Molecular Systems Design & Engineering · doi:10.1039/c9me00039a · Cited by 683 · cited 1× · see citing works
“Although these models vary greatly in method [5], they can usually be categorised by four common architectures/approaches: (1) Language models (such as [6–8]) which require a chemical language (e.”
MethodGraph Convolutional Policy Network for Goal-Directed Molecular Graph GenerationPreprint
You, J., Liu, B., Ying, R., Pande, V. S., & Leskovec, J. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1806.02473 · Cited by 644 · see citing works
A Comprehensive Listing of Bioactivation Pathways of Organic Functional Groups
Kalgutkar, A. S. S., Gardner, I. B., Obach, R. S., Shaffer, C. L., Callegari, E., Henne, K. R., Mutlib, A. E., Dalvie, D., Lee, J. S., Nakai, Y., et al. (2005) · Current Drug Metabolism · doi:10.2174/1389200054021799 · Cited by 623 · cited 1× · see citing works
“5; no charged atoms; no atoms besides C, N, S, O, F, Cl, Br, H; no cycles larger than 8 members; custom medicinal chemistry filters [61, 62]; and finally PAINS filters [63] were applied.”
BackgroundSelf-referencing embedded strings (SELFIES): A 100% robust molecular string representation
Krenn, M., Häse, F., Nigam, A., Friederich, P., & Aspuru‐Guzik, A. (2020) · Repository KITopen (Karlsruhe Institute of Technology) · doi:10.5445/ir/1000128111 · Cited by 616 · see citing works
Optimization of Molecules via Deep Reinforcement Learning
Zhou, Z., Kearnes, S. M., Li, L., Zare, R. N., & Riley, P. F. (2019) · Scientific Reports · doi:10.1038/s41598-019-47148-x · Cited by 602 · see citing works
Large-scale comparison of machine learning methods for drug target prediction on ChEMBL
Mayr, A., Klambauer, G., Unterthiner, T., Steijaert, M., Wegner, J. K., Ceulemans, H., Clevert, D.-A., & Hochreiter, S. (2018) · Chemical Science · doi:10.1039/c8sc00148k · Cited by 584 · cited 2× · see citing works
“11 • Fréchet ChemNet Distance (FCD) [79] was used to enable comparison with previous studies, which measures the mean and covariance of the penultimate layer of ChemNet [80] for two datasets.”
MethodDeng, Z., Chuaqui, C., & Singh, J. (2004) · Journal of Medicinal Chemistry · doi:10.1021/jm030331x · Cited by 544 · see citing works
Molecular Shape Diversity of Combinatorial Libraries: A Prerequisite for Broad Bioactivity
Sauer, W., & Schwarz, M. K. (2003) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci025599w · Cited by 500 · cited 4× · see citing works
“Furthermore, we investigate the 3D shape of molecules by looking at the normalized principal moments ratio (NPR) [90].”
MethodInfluential‘It will change everything’: DeepMind’s AI makes gigantic leap in solving protein structures
Callaway, E. (2020) · Nature · doi:10.1038/d41586-020-03348-4 · Cited by 499 · cited 1× · see citing works
“As structural input, either experimentally resolved structures or homology models can be employed and given the increasing numbers of structures available (which increases by about 10,000 per year [40]) and development of protein structure prediction technology [41], this renders this approach applicable to an increas…”
BackgroundStructure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone
Wang, S., Che, T., Levit, A., Shoichet, B. K., Wacker, D., & Roth, B. L. (2018) · Nature · doi:10.1038/nature25758 · Cited by 487 · cited 1× · see citing works
“DRD2 also has a publicly available X-ray crystal structure [45] in complex with Risperidone, thereby allowing use of molecular docking without the requirement of generating a homology model.”
BackgroundMolGAN: An implicit generative model for small molecular graphs
De Cao, N., & Kipf, T. (2018) · arXiv · doi:10.48550/arxiv.1805.11973 · Cited by 442 · see citing works
A New Atom-Additive Method for Calculating Partition Coefficients
Wang, R., Fu, Y., & Lai, L. (1997) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci960169p · Cited by 426 · cited 2× · see citing works
“In short, molecules were selected to adhere to the following rules: molecular weight between 250–350 Da; number of rotatable bonds not greater than 8; XlogP [60] not greater than 3.”
BackgroundInfluentialMéndez‐Lucio, O., Baillif, B., Clevert, D.-A., Rouquié, D., & Wichard, J. (2020) · Nature Communications · doi:10.1038/s41467-019-13807-w · Cited by 422 · see citing works
The Synthesizability of Molecules Proposed by Generative Models
Gao, W., & Coley, C. W. (2020) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c00174 · Cited by 405 · cited 1× · see citing works
“Despite the fact that goal-directed optimization tasks have previously led to worse syntheizability [100].”
BackgroundA de novo molecular generation method using latent vector based generative adversarial network
Prykhodko, O., Johansson, S., Kotsias, P.-C., Arús‐Pous, J., Bjerrum, E. J., Engkvist, O., & Chen, H. (2019) · Journal of Cheminformatics · doi:10.1186/s13321-019-0397-9 · Cited by 395 · cited 1× · see citing works
“biased training or fine-tuning) [2, 6, 19], or using machine learning models trained on known bioactive molecules to predict de novo molecule bioactivity (e.”
Impact of GPCR Structures on Drug Discovery
Congreve, M., de Graaf, C., Swain, N. A., & Tate, C. G. (2020) · Cell · doi:10.1016/j.cell.2020.03.003 · Cited by 380 · cited 1× · see citing works
“Although more structures are released every year [40], which offers an ever increasing opportunity to utilise structure-based design [47].”
BackgroundGrammar Variational AutoencoderPreprint
Kusner, M. J., Paige, B., & Hernández-Lobato, J. M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.01925 · Cited by 367 · cited 1× · see citing works
“[52] evidenced that CVAE [10] and GVAE [53] were unable to generate molecules with optimized Smina [43] docking scores due to the inaccurate prediction of said docking score, which is used to guide de novo sampling in the respective methods.”
MethodFréchet ChemNet Distance: A Metric for Generative Models for Molecules in Drug Discovery
Preuer, K., Renz, P., Unterthiner, T., Hochreiter, S., & Klambauer, G. (2018) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.8b00234 · Cited by 345 · cited 3× · see citing works
“Abbreviations ADAM: Adaptative moment estimation; DRD2: Dopamine receptor D2; FEP: Free energy perturbation; FCD: Fréchet chemnet distance; GPCR: G protein-coupled receptor; QED: Quantitative estimate of drug-likeness; QSAR: Quantitative structure–activity relationship; SAscore: Synthetic accessibility score; SELFIES:…”
BackgroundResultSimilarity to Molecules in the Training Set Is a Good Discriminator for Prediction Accuracy in QSAR
Sheridan, R. P., Feuston, B. P., Maiorov, V. N., & Kearsley, S. K. (2004) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci049782w · Cited by 298 · cited 1× · see citing works
“This means that models will score molecules similar to those observed in the training data more accurately [26].”
MethodMost Ligand-Based Classification Benchmarks Reward Memorization Rather than Generalization
Wallach, I., & Heifets, A. (2018) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.7b00403 · Cited by 287 · cited 1× · see citing works
“they perform well on ‘in-distribution’ data but struggle to extrapolate to ‘out-of-distribution’ data, which is often poorly accounted for in model validation [24, 25].”
BackgroundAssessing the impact of generative AI on medicinal chemistry
Walters, W. P., & Murcko, M. (2020) · Nature Biotechnology · doi:10.1038/s41587-020-0418-2 · Cited by 262 · cited 1× · see citing works
“This lack of novelty observed in current deep generative models has also been commented on in the literature [30].”
BackgroundEfficient multi-objective molecular optimization in a continuous latent space
Winter, R., Montanari, F., Steffen, A., Briem, H., Noé, F., & Clevert, D.-A. (2019) · Chemical Science · doi:10.1039/c9sc01928f · Cited by 261 · cited 1× · see citing works
“using reinforcement learning [7, 8, 20, 21], Bayesian optimization [22] or particle swarm optimization [23].”
MethodDeep Reinforcement Learning for Multiparameter Optimization in de novo Drug Design
Ståhl, N., Falkman, G., Karlsson, A., Mathiason, G., & Boström, J. (2019) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.9b00325 · Cited by 239 · see citing works
Kotsias, P.-C., Arús‐Pous, J., Chen, H., Engkvist, O., Tyrchan, C., & Bjerrum, E. J. (2020) · Nature Machine Intelligence · doi:10.1038/s42256-020-0174-5 · Cited by 235 · cited 1× · see citing works
“This receptor has a wealth of associated ligand bioactivity data available, and it has been commonly used in deep generative model publications before [7, 21, 22, 29, 44], thereby allowing any further comparison to different methods.”
MethodMinimising the potential for metabolic activation in drug discovery
Kalgutkar, A. S. S., & Soglia, J. R. (2005) · Expert Opinion on Drug Metabolism & Toxicology · doi:10.1517/17425255.1.1.91 · Cited by 233 · cited 1× · see citing works
“5; no charged atoms; no atoms besides C, N, S, O, F, Cl, Br, H; no cycles larger than 8 members; custom medicinal chemistry filters [61, 62]; and finally PAINS filters [63] were applied.”
BackgroundExCAPE-DB: an integrated large scale dataset facilitating Big Data analysis in chemogenomics
Sun, J., Jeliazkova, N., Chupakhin, V., Dzib, J. F. G., Engkvist, O., Carlsson, L., Wegner, J. K., Ceulemans, H., Georgiev, I., Jeliazkov, V., et al. (2017) · Journal of Cheminformatics · doi:10.1186/s13321-017-0203-5 · Cited by 227 · cited 6× · see citing works
“We (1) first removed known DRD2 actives (according to the ExCAPE-DB [58]) from the MOSES curated [4] ZINC [59] database of small drug-like molecules for use as training data.”
BackgroundMethodInfluentialBianchi, F., Klooster, J. S. V. ’., Ruiz, S. J., Luck, K., Pols, T., Urbatsch, I. L., & Poolman, B. (2016) · Scientific Reports · doi:10.1038/srep31443 · Cited by 219 · see citing works
Exploring the GDB-13 chemical space using deep generative models
Arús‐Pous, J., Blaschke, T., Ulander, S., Reymond, J., Chen, H., & Engkvist, O. (2019) · Journal of Cheminformatics · doi:10.1186/s13321-019-0341-z · Cited by 191 · cited 1× · see citing works
“docking scores and as such variable size distributions can be sampled (up to billions [57]) of molecules, as opposed to sampling a finite latent space as in [51].”
BackgroundDixon, S. L., Duan, J., Smith, E., Bargen, C. D. V., Sherman, W., & Repasky, M. P. (2016) · Future Medicinal Chemistry · doi:10.4155/fmc-2016-0093 · Cited by 185 · cited 1× · see citing works
“To achieve this, they trained an AutoQSAR model [49] on a subset of 1,000 enumerated analogues of a potent inhibitor with the corresponding FEP predictions, which was subsequently used as the REINVENT scoring function.”
MethodPan, Y., Huang, N., Cho, S., & MacKerell, A. D. (2002) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci020055f · Cited by 157 · cited 1× · see citing works
“This is in contrast to previous publications, which frequently found that larger molecules are favoured by many scoring functions [103, 104].”
ResultOn failure modes in molecule generation and optimization
Renz, P., Rompaey, D. V., Wegner, J. K., Hochreiter, S., & Klambauer, G. (2019) · Drug Discovery Today Technologies · doi:10.1016/j.ddtec.2020.09.003 · Cited by 154 · cited 2× · see citing works
“observed different chemical space occupation for independent runs with similar models [27].”
Sánchez-Lengeling, B., Outeiral, C., Guimaraes, G. L., & Aspuru‐Guzik, A. (2017) · ChemRxiv · doi:10.26434/chemrxiv.5309668.v2 · Cited by 148 · see citing works
Memory-assisted reinforcement learning for diverse molecular de novo design
Blaschke, T., Engkvist, O., Bajorath, J., & Chen, H. (2020) · Journal of Cheminformatics · doi:10.1186/s13321-020-00473-0 · Cited by 130 · cited 6× · see citing works
“This is very likely a reason for the lack of diversity (and inability to access truly novel chemical space) seen in deep generative models [28, 29].”
BackgroundMethodResultInfluentialLearning Multimodal Graph-to-Graph Translation for Molecular OptimizationPreprint
Jin, W., Yang, K., Barzilay, R., & Jaakkola, T. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.01070 · Cited by 129 · cited 2× · see citing works
“This receptor has a wealth of associated ligand bioactivity data available, and it has been commonly used in deep generative model publications before [7, 21, 22, 29, 43], thereby allowing any further comparison to different methods.”
MethodChemGAN challenge for drug discovery: can AI reproduce natural chemical diversity?Preprint
Benhenda, M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1708.08227 · Cited by 108 · cited 1× · see citing works
“This increase in FCD has shown to indicate a number of differences [79] to the Prior training data for example, ‘drug-likeness’ defined by [94] or internal diversity [98].”
ResultPredicting the accuracy of protein–ligand docking on homology models
Bordogna, A., Pandini, A., & Bonati, L. (2011) · Journal of Computational Chemistry · doi:10.1002/jcc.21601 · Cited by 107 · cited 1× · see citing works
“Provided a scoring function achieves enrichment of bioactive compounds against a protein target (which can be established on existing datasets where data is available, but where otherwise estimates can be made based on the character of the binding pocket and protein type [38, 39]), then the chemical space to be scored…”
BackgroundCan we use docking and scoring for hit-to-lead optimization?
Enyedy, I. J., & Egan, W. J. (2008) · Journal of Computer-Aided Molecular Design · doi:10.1007/s10822-007-9165-4 · Cited by 97 · cited 1× · see citing works
“4 inaccurate [35, 36] and the performance of scoring functions can be highly target-dependent [37], molecular docking consistently results in the early enrichment of known active molecules in virtual libraries compared to random [35] and is a generally-applied computational ligand design method in pharmaceutical resea…”
BackgroundFunction-specific virtual screening for GPCR ligands using a combined scoring method
Kooistra, A. J., Vischer, H. F., McNaught-Flores, D. A., Leurs, R., de Esch, I. J. P., & de Graaf, C. (2016) · Scientific Reports · doi:10.1038/srep28288 · Cited by 90 · cited 1× · see citing works
“As expected, known DRD2 ligands form a hydrogen-bond interaction with D1143x32, a highly conserved residue in aminergic receptors that has been shown to be crucial for ligand binding [66, 67].”
BackgroundLearning To Navigate The Synthetically Accessible Chemical Space Using Reinforcement LearningPreprint
Gottipati, S. K., Sattarov, B., Niu, S., Pathak, Y., Wei, H., Liu, S., Thomas, K. M. J., Blackburn, S., Coley, C. W., Tang, J., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.12485 · Cited by 82 · cited 1× · see citing works
“using reinforcement learning [7, 8, 20, 21], Bayesian optimization [22] or particle swarm optimization”
MethodAmabilino, S., Pogány, P., Pickett, S. D., & Green, D. V. S. (2020) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c00343 · Cited by 74 · cited 1× · see citing works
“This is very likely a reason for the lack of diversity (and inability to access truly novel chemical space) seen in deep generative models [28, 29].”
BackgroundDixon, S. L., & Koehler, R. T. (1999) · Journal of Medicinal Chemistry · doi:10.1021/jm980708c · Cited by 73 · cited 3× · see citing works
“Most notably, internal diversity measures GDB13 as more diverse than GDB17—which contradicts chemical intuition, but further confers with hypothesized confounding effects [81].”
BackgroundInfluentialCleves, A. E., & Jain, A. N. (2007) · Journal of Computer-Aided Molecular Design · doi:10.1007/s10822-007-9150-y · Cited by 70 · cited 1× · see citing works
“they perform well on ‘in-distribution’ data but struggle to extrapolate to ‘out-of-distribution’ data, which is often poorly accounted for in model validation [24, 25].”
BackgroundGenerative Models Should at Least Be Able to Design Molecules That Dock Well: A New Benchmark
Ciepliński, T., Danel, T., Podlewska, S., et al. (2023) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.2c01355 · Cited by 69 · see citing works
Li, X., Xu, Y., Yao, H., & Lin, K. (2020) · Journal of Cheminformatics · doi:10.1186/s13321-020-00446-3 · Cited by 68 · cited 2× · see citing works
“[50] trained a recurrent neural network on known kinase CDK4 inhibitors and finetuned the network by training on a selection of generated molecules screened using docking.”
BackgroundResultOptiMol : Optimization of Binding Affinities in Chemical Space for Drug Discovery
Boitreaud, J., Mallet, V., Oliver, C., & Waldispühl, J. (2020) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c00833 · Cited by 67 · see citing works
Molecular Design in Synthetically Accessible Chemical Space via Deep Reinforcement Learning
Horwood, J., & Noutahi, E. (2020) · ACS Omega · doi:10.1021/acsomega.0c04153 · Cited by 65 · cited 2× · see citing works
“using reinforcement learning [7, 8, 20, 21], Bayesian optimization [22] or particle swarm optimization [23].”
MethodMolecular interaction fingerprint approaches for GPCR drug discovery
Vass, M., Kooistra, A. J., Ritschel, T., Leurs, R., de Esch, I. J. P., & de Graaf, C. (2016) · Current Opinion in Pharmacology · doi:10.1016/j.coph.2016.07.007 · Cited by 64 · see citing works
Comparative Study of Deep Generative Models on Chemical Space Coverage
Zhang, J., Mercado, R., Engkvist, O., & Chen, H. (2021) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c01328 · Cited by 63 · see citing works
DISE: Directed Sphere Exclusion
Gobbi, A., & Lee, M. (2002) · Journal of Chemical Information and Computer Sciences · doi:10.1021/ci025554v · Cited by 62 · see citing works
Structure‐Based Virtual Screening for Dopamine D 2 Receptor Ligands as Potential Antipsychotics
Kaczor, A. A., Silva, A., Loza, M. I., Kolb, P., Castro, M., & Poso, A. (2016) · ChemMedChem · doi:10.1002/cmdc.201500599 · Cited by 62 · see citing works
Can easy chemistry produce complex, diverse, and novel molecules?
Tomberg, A., & Boström, J. (2020) · Drug Discovery Today · doi:10.1016/j.drudis.2020.09.027 · Cited by 59 · see citing works
Unbiasing Scoring Functions: A New Normalization and Rescoring Strategy
Carta, G., Knox, A. J. S., & Lloyd, D. G. (2007) · Journal of Chemical Information and Modeling · doi:10.1021/ci600471m · Cited by 56 · see citing works
Ghanakota, P., Bos, P. H., Konze, K. D., Staker, J., Marques, G. F. F., Marshall, K., Leswing, K., Abel, R., & Bhat, S. (2020) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c00120 · Cited by 50 · see citing works
Krämer, C., Fuchs, J. E., & Liedl, K. R. (2015) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.5b00018 · Cited by 48 · see citing works
Structure-Based Predictions of Activity Cliffs
Husby, J., Bottegoni, G., Kufareva, I., Abagyan, R. A., & Cavalli, A. (2015) · Journal of Chemical Information and Modeling · doi:10.1021/ci500742b · Cited by 40 · see citing works
Has Drug Design Augmented by Artificial Intelligence Become a Reality?
Chen, H., & Engkvist, O. (2019) · Trends in Pharmacological Sciences · doi:10.1016/j.tips.2019.09.004 · Cited by 40 · see citing works
Du, H., Brender, J., Zhang, J., & Zhang, Y. (2014) · Methods · doi:10.1016/j.ymeth.2014.08.017 · Cited by 38 · cited 1× · see citing works
“Provided a scoring function achieves enrichment of bioactive compounds against a protein target (which can be established on existing datasets where data is available, but where otherwise estimates can be made based on the character of the binding pocket and protein type [38, 39]), then the chemical space to be scored…”
BackgroundUsing a genetic algorithm to find molecules with good docking scores
Steinmann, C., & Jensen, J. H. (2021) · PeerJ Physical Chemistry · doi:10.7717/peerj-pchem.18 · Cited by 38 · see citing works
Primary structure of theLeishmania donovanismall subunit ribosomal RNA coding region
Looker, D. L., Miller, L. A., Elwood, H. J., et al. (1988) · Nucleic Acids Research · doi:10.1093/nar/16.14.7198 · Cited by 34 · see citing works
Discovery, Optimization, and Characterization of Novel D2 Dopamine Receptor Selective Antagonists
Xiao, J., Free, R. B., Barnaeva, E. S., et al. (2014) · Journal of Medicinal Chemistry · doi:10.1021/jm500126s · Cited by 30 · cited 1× · see citing works
“[97] which had an experimental hit rate against DRD2 of ~0.”
BackgroundNavigating Chemical Space by Interfacing Generative Artificial Intelligence and Molecular Docking
Xu, Z., Wauchope, O. R., & Frank, A. T. (2021) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.1c00746 · Cited by 24 · see citing works
The Synthesizability of Molecules Proposed by Generative ModelsPreprint
Gao, W., & Coley, C. W. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.07007 · Cited by 8 · cited 1× · see citing works
“Despite the fact that goal-directed optimization tasks have previously led to worse syntheizability [100].”
BackgroundUsing a Genetic Algorithm to Find Molecules with Good Docking ScoresPreprint
Steinmann, C., & Jensen, J. H. (2021) · ChemRxiv · doi:10.26434/chemrxiv.13525589 · Cited by 4 · see citing works
(2021) · doi:10.6019/chembl.database.28 · Cited by 3 · see citing works
Flagged first, then by citations
2 of the 105 references couldn’t be matched to a known paper.
Research works printed but not linked: 3
- Cieplinski T, Danel T, Podlewska S, Jastrzębski S. We should at least be able to design molecules that dock well. arXiv:2006.16955 [q-bio.BM] · arXiv
- Kusner MJ, Paige B, Miguel Hernández-Lobato J. Grammar variational autoencoder. arXiv:1703.01925 [stat.ML] · arXiv
- Danel T, Szymczak M, Maziarka Ł, Podolak I, Tabor J, Jastrz˛ S. De Novo Drug Design with a Docking Score Proxy. In: Machine Learning for Molecules Workshop at NeurIPS 2020. 2020
Cited by
Sorted by citations
Han, R., Yoon, H., Kim, G., Lee, H., & Lee, Y. (2023) · Pharmaceuticals · doi:10.3390/ph16091259 · Cited by 192 · see citing works
“It requires calculating pairwise similarities for all data points, which can be computationally demanding for large datasets [55].”
BackgroundGenerative machine learning for de novo drug discovery: A systematic review
Martinelli, D. D. (2022) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2022.105403 · Cited by 177 · see citing works
Application of Computational Biology and Artificial Intelligence in Drug Design
Zhang, Y., Luo, M., Wu, P., Wu, S., Lee, T.-Y., & Bai, C. (2022) · International Journal of Molecular Sciences · doi:10.3390/ijms232113568 · Cited by 160 · see citing works
DOCKSTRING: Easy Molecular Docking Yields Better Benchmarks for Ligand Design
García-Ortegón, M., Simm, G. N. C., Tripp, A., Hernández-Lobato, J. M., Bender, A., & Bacallado, S. (2022) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.1c01334 · Cited by 137 · see citing works
Self-citationLibINVENT: Reaction-based Generative Scaffold Decoration for in Silico Library Design
Fialková, V., Zhao, J., Papadopoulos, K., Engkvist, O., Bjerrum, E. J., Kogej, T., & Patronov, A. (2021) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.1c00469 · Cited by 107 · see citing works
Link-INVENT: Generative Linker Design with Reinforcement Learning
Guo, J., Knuth, F., Margreitter, C., Janet, J. P., Papadopoulos, K., Engkvist, O., & Patronov, A. (2022) · doi:10.26434/chemrxiv-2022-qkx9f · Cited by 101 · see citing works
Papyrus: a large-scale curated dataset aimed at bioactivity predictions
Béquignon, O. J. M., Bongers, B. J., Jespers, W., IJzerman, A. P., van der Water, B., & van Westen, G. J. P. (2023) · Journal of Cheminformatics · doi:10.1186/s13321-022-00672-x · Cited by 83 · see citing works
“The authors interpret this as the minimum fraction of the dataset required to explain the chemical diversity in the context of bioactivity [35].”
Deep scaffold hopping with multimodal transformer neural networks
Zheng, S., Lei, Z., Ai, H., Chen, H., Deng, D., & Yang, Y. (2021) · Journal of Cheminformatics · doi:10.1186/s13321-021-00565-5 · Cited by 81 · see citing works
DockStream: a docking wrapper to enhance de novo molecular design
Guo, J., Janet, J. P., Bauer, M. R., Nittinger, E., Giblin, K. A., Papadopoulos, K., Voronov, A., Patronov, A., Engkvist, O., & Margreitter, C. (2021) · Journal of Cheminformatics · doi:10.1186/s13321-021-00563-7 · Cited by 77 · see citing works
“18 – 21 However, it is often challenging to ascertain what exactly constitutes a ‘good’ docking score.”
BackgroundGenerative Models Should at Least Be Able to Design Molecules That Dock Well: A New Benchmark
Ciepliński, T., Danel, T., Podlewska, S., et al. (2023) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.2c01355 · Cited by 69 · see citing works
Sahu, M. K., Nayak, A. K., Hailemeskel, B., et al. (2024) · International Journal Of Pharmaceutical Research And Allied Sciences · doi:10.51847/une9jqjucl · Cited by 67 · see citing works
“By exploring chemical space and identifying analogs with similar pharmacological profiles, ligand-based docking aids in the design of novel ligands with improved efficacy and reduced off-target effects [54].”
BackgroundImproving de novo molecular design with curriculum learning
Guo, J., Fialková, V., Arango, J. D., Margreitter, C., Janet, J. P., Papadopoulos, K., Engkvist, O., & Patronov, A. (2022) · Nature Machine Intelligence · doi:10.1038/s42256-022-00494-4 · Cited by 59 · see citing works
“…we demonstrate that simple curricula, utilizing a single curriculum objective, can accelerate agent productivity and generate compounds that satisfy a docking constraint, that is, predicted to retain experimentally validated interactions (see Methods for experiment hyperparameters) 9,10,16–18 .”
BackgroundIntegrating structure-based approaches in generative molecular design
Thomas, M. C., Bender, A., & de Graaf, C. (2023) · Current Opinion in Structural Biology · doi:10.1016/j.sbi.2023.102559 · Cited by 58 · see citing works
Self-citationThomas, M. C., O’Boyle, N. M., Bender, A., & de Graaf, C. (2022) · Journal of Cheminformatics · doi:10.1186/s13321-022-00646-z · Cited by 55 · see citing works
“Specifically a RNN was trained on the MOSESn dataset [23, 52], an agent was initialized which then underwent RL updates to optimize the docking score of de novo molecules against DRD2.”
BackgroundMethodResultInfluentialSelf-citationAdvancing Targeted Protein Degradation via Multiomics Profiling and Artificial Intelligence
Duran-Frigola, M., Cigler, M., & Winter, G. E. (2023) · Journal of the American Chemical Society · doi:10.1021/jacs.2c11098 · Cited by 50 · see citing works
Deep generative models for 3D molecular structure
Baillif, B., Cole, J. C., McCabe, P., & Bender, A. (2023) · Current Opinion in Structural Biology · doi:10.1016/j.sbi.2023.102566 · Cited by 46 · see citing works
Self-citationDocking-based generative approaches in the search for new drug candidates
Danel, T., Łȩski, J., Podlewska, S., & Podolak, I. T. (2022) · Drug Discovery Today · doi:10.1016/j.drudis.2022.103439 · Cited by 45 · see citing works
The application of artificial intelligence to accelerate G protein‐coupled receptor drug discovery
Nguyen, A. T. N., Nguyen, D. T. N., Koh, H. Y., Toskov, J., MacLean, W., Xu, A., Zhang, D., Webb, G. I., May, L. T., & Halls, M. L. (2023) · British Journal of Pharmacology · doi:10.1111/bph.16140 · Cited by 42 · see citing works
A systematic review of deep learning chemical language models in recent era
Flores-Hernandez, H., & Martínez-Ledesma, E. (2024) · Journal of Cheminformatics · doi:10.1186/s13321-024-00916-y · Cited by 46 · see citing works
Real-World Applications and Experiences of AI/ML Deployment for Drug Discovery
Pitt, W. R., Bentley, J. M., Boldron, C., Colliandre, L., Esposito, C., Frush, E. H., Kopec, J., Labouille, S., Meneyrol, J., Pardoe, D. A., et al. (2025) · Journal of Medicinal Chemistry · doi:10.1021/acs.jmedchem.4c03044 · Cited by 43 · see citing works
New strategies to enhance the efficiency and precision of drug discovery
An, Q., Huang, L., Wang, C., Wang, D., & Tu, Y. (2025) · Frontiers in Pharmacology · doi:10.3389/fphar.2025.1550158 · Cited by 39 · see citing works
Thomas, M. C., O’Boyle, N. M., Bender, A., & de Graaf, C. (2024) · Journal of Cheminformatics · doi:10.1186/s13321-024-00861-w · Cited by 39 · see citing works
Self-citationSaifi, I., Bhat, B. A., Hamdani, S. S., Bhat, U. Y., Lobato-Tapia, C. A., Mir, M. A., Dar, T. U. H., & Ganie, S. A. (2023) · Journal of Biomolecular Structure and Dynamics · doi:10.1080/07391102.2023.2234039 · Cited by 36 · see citing works
ACEGEN: Reinforcement Learning of Generative Chemical Agents for Drug Discovery
Bou, A., Thomas, M., Dittert, S., Navarro, C., Majewski, M., Wang, Y., Patel, S., Tresadern, G., Ahmad, M., Moens, V., et al. (2024) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.4c00895 · Cited by 38 · see citing works
“More-over, we use sphere exclusion diversity (SEDiv@1k) of de novo molecules Thomas et al. (2021) which measures the proportion of molecules needed to describe chemical space in a random sample of 1,000 molecules, as a proxy for exploration.”
Self-citationTosstorff, A., Rudolph, M. G., Cole, J. C., Reutlinger, M., Krämer, C., Schaffhauser, H., Nilly, A., Flohr, A., & Kuhn, B. (2022) · Journal of Computer-Aided Molecular Design · doi:10.1007/s10822-022-00478-x · Cited by 34 · see citing works
“…parent compound improved ranking performance, the docking workflow presented here can lend itself to additional applications, such as the generation of starting poses for free-energy perturbation calculations or as a target function in generative neural networks, as has been described before [21].”
BackgroundExplaining and avoiding failure modes in goal-directed generation of small molecules
Langevin, M., Vuilleumier, R., & Bianciotto, M. (2022) · Journal of Cheminformatics · doi:10.1186/s13321-022-00601-y · Cited by 32 · see citing works
“2 and “Methods” section for further details) and sometimes Those results, suggesting that the molecules produced through goal-directed generation exploit bias unique to the model they are optimized on, were noted in the literature [16, 17].”
BackgroundThe Application of Artificial Intelligence Accelerates G Protein-Coupled Receptor Ligand Discovery
Chen, W., Song, C., Leng, L., Zhang, S., & Chen, S. (2023) · Engineering · doi:10.1016/j.eng.2023.09.011 · Cited by 28 · see citing works
D, K. H. M., Thomas, M. C., Kageyama, M., et al. (2023) · Journal of Cheminformatics · doi:10.1186/s13321-023-00781-1 · Cited by 26 · see citing works
“…model) by calculating the average similarity of generated molecules to the single nearest neighbour (aSNN) present in a given part of the dataset, based on the Tanimoto similarity of Morgan fingerprints using RDKit [12, 48] and with the aSNN calculations performed as implemented in MolScore [39].”
MethodSelf-citationThomas, M. C., Matricon, P. G., Gillespie, R. F., Napiórkowska, M., Neale, H., Mason, J. S., Brown, J., Harwood, K., Fieldhouse, C., Swain, N. A., et al. (2025) · Nature Communications · doi:10.1038/s41467-025-60629-0 · Cited by 25 · see citing works
Self-citationNavigating Chemical Space by Interfacing Generative Artificial Intelligence and Molecular Docking
Xu, Z., Wauchope, O. R., & Frank, A. T. (2021) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.1c00746 · Cited by 24 · see citing works
Hit and Lead Discovery with Explorative RL and Fragment-based Molecule GenerationPreprint
Yang, S., Hwang, D., Lee, S., Ryu, S., & Hwang, S. J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.01219 · Cited by 24 · see citing works
AI meets physics in computational structure-based drug discovery for GPCRs
Michino, M., Vendôme, J., & Kufareva, I. (2025) · npj Drug Discovery. · doi:10.1038/s44386-025-00019-0 · Cited by 25 · see citing works
Comprehensive machine learning boosts structure-based virtual screening for PARP1 inhibitors
Caba, K., Tran‐Nguyen, V., Rahman, T., & Ballester, P. J. (2024) · Journal of Cheminformatics · doi:10.1186/s13321-024-00832-1 · Cited by 21 · see citing works
“This structure-based approach was motivated by comprehensive comparative studies across targets showing that models exploiting protein-ligand features outperform those based on ligand-only features [68,70,71].”
BackgroundDiverse Hits in De Novo Molecule Design: Diversity-Based Comparison of Goal-Directed Generators
Renz, P., Luukkonen, S., & Klambauer, G. (2024) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.4c00519 · Cited by 18 · see citing works
Activity cliff-aware reinforcement learning for de novo drug design
Hu, X., Liu, G., Zhao, Y., et al. (2025) · Journal of Cheminformatics · doi:10.1186/s13321-025-01006-3 · Cited by 18 · see citing works
“In contrast, structure-based docking software has been proven to reflect activity cliffs authentically [22, 23], leading to calls for the use of docking in the evaluation of drug design algorithms, as opposed to simpler scoring functions [24, 25].”
Benchmarking 3D Structure-Based Molecule Generators
Sanjrani, N., Coupry, D. E., Pogány, P., Palmer, D. S., & Pickett, S. D. (2025) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.5c01020 · Cited by 17 · see citing works
A Simple Way to Incorporate Target Structural Information in Molecular Generative Models
Zhang, W., Zhang, K., & Huang, J. (2023) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.3c00293 · Cited by 14 · see citing works
Sauer, S., Matter, H., Heßler, G., & Grebner, C. (2022) · Frontiers in Chemistry · doi:10.3389/fchem.2022.1012507 · Cited by 14 · see citing works
No dance, no partner! A tale of receptor flexibility in docking and virtual screening
Basciu, A., Callea, L., Motta, S., Bonvin, A. M. J. J., Bonati, L., & Vargiu, A. V. (2022) · Annual reports in medicinal chemistry · doi:10.1016/bs.armc.2022.08.006 · Cited by 14 · see citing works
Application of computational methods for class A GPCR Ligand discovery
Szwabowski, G. L., Baker, D. L., & Parrill, A. L. (2023) · Journal of Molecular Graphics and Modelling · doi:10.1016/j.jmgm.2023.108434 · Cited by 13 · see citing works
Bosc, N., Félix, E., Gardner, M., Mills, J. E. J., Timmerman, M., Asveld, D., Rensen, K., Mukherjee, P., Das, R., Chenu, E., et al. (2023) · ACS Medicinal Chemistry Letters · doi:10.1021/acsmedchemlett.3c00369 · Cited by 8 · see citing works
Decrypting orphan GPCR drug discovery via multitask learning
Huang, W., Lin, W.-T., Hung, M., Lee, J., & Tung, C.-W. (2024) · Journal of Cheminformatics · doi:10.1186/s13321-024-00806-3 · Cited by 7 · see citing works
Xiong, Y., Wang, Y., Wang, Y., Li, C., Yusong, P., Wu, J., Wang, Y., Gu, L., & Butch, C. J. (2023) · Journal of Computer-Aided Molecular Design · doi:10.1007/s10822-023-00523-3 · Cited by 7 · see citing works
Lib-INVENT: Reaction Based Generative Scaffold Decoration for in silico Library DesignPreprint
Fialková, V., Zhao, J., Papadopoulos, K., Engkvist, O., Bjerrum, E. J., Kogej, T., & Patronov, A. (2021) · doi:10.33774/chemrxiv-2021-57sv1-v2 · Cited by 7 · see citing works
Test-Time Training Scaling Laws for Chemical Exploration in Drug Design
Thomas, M. C., Bou, A., & Fabritiis, G. D. (2025) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.5c02316 · Cited by 6 · see citing works
“Molecular diversity is also measured by sphere exclusion diversity (SEDiv) at a sample size of 1000 44 which provides a more accurate representation of chemical space coverage than internal diversity.”
Self-citationDockStream: A Docking Wrapper to Enhance De Novo Molecular DesignPreprint
Guo, J., Janet, J. P., Bauer, M. R., Nittinger, E., Giblin, K. A., Papadopoulos, K., Voronov, A., Patronov, A., Engkvist, O., & Margreitter, C. (2021) · doi:10.33774/chemrxiv-2021-qvhml · Cited by 6 · see citing works
Integrated Computer-Aided Drug Design: Advances in GPCR Natural Ligand Discovery
Uba, A. İ. (2025) · Cell Biochemistry and Biophysics · doi:10.1007/s12013-025-01902-3 · Cited by 5 · see citing works
Metcalf, D. P., Glick, Z. L., Bortolato, A., Jiang, A., Cheney, D. L., & Sherrill, C. D. (2024) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.3c02054 · Cited by 5 · see citing works
Lim, H. (2024) · Journal of Cheminformatics · doi:10.1186/s13321-024-00845-w · Cited by 6 · see citing works
“I employed the commonly used metrics in de novo design (validity, uniqueness, novelty, and internal diversity) to compare the pre-trained models, but it’s worth noting that internal diversity cannot capture all aspects of a compound’s diversity [63].”
MethodLim, H. (2024) · Materials Today Advances · doi:10.1016/j.mtadv.2024.100529 · Cited by 5 · see citing works
Armando, R. G., Cabrera, M., Vilarullo, R., Chinestrad, P., Maggio, J., Paderta, C., Menna, P. L., Gomez, D. E., & Gómez, D. L. M. (2022) · Oncology Reports · doi:10.3892/or.2022.8403 · Cited by 5 · see citing works
Navigating Ultralarge Virtual Chemical Spaces with Product-of-Experts Chemical Language Models
Nakata, S., Mori, Y., & Tanaka, S. (2024) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.4c01214 · Cited by 4 · see citing works
Wang, Y., Wang, Z., Li, Y., et al. (2025) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.5c00944 · Cited by 4 · see citing works
SAGE-Amine: Generative Amine design with multi-property optimization for efficient CO2 capture
Lim, H., Cho, H., Kim, J.-H., & No, K. T. (2025) · Carbon Capture Science & Technology · doi:10.1016/j.ccst.2025.100447 · Cited by 3 · see citing works
“Diversity is evaluated by analyzing the chemical variation within the generated molecule sets, determined using sphere exclusion diversity (SEDiv [48]).”
Artificial intelligence-assisted identification and screening strategies in sweetener design
Xie, H., Jiao, Q., Li, H., Ye, H. L., Ren, G., Huang, M. Z., & Yang, T. (2025) · Critical Reviews in Food Science and Nutrition · doi:10.1080/10408398.2025.2516136 · Cited by 3 · see citing works
“Previous findings indicate that structure-based methodologies, especially those augmented by AI, can effectively overcome certain limitations associated with ligand-based scoring functions (Zhang et al. 2024; Zhang et al. 2023; Thomas et al. 2021).”
A comprehensive review of chemoinformatics: From molecular representation to data-driven design
Islam, M. S., Sarker, P., Islam, M. M., Shamim, S. U., Malitha, S. B., Nakib, K. I. A., Alam, M. Z., & Chowdhury, A. M. S. (2026) · Next research. · doi:10.1016/j.nexres.2026.102362 · Cited by 2 · see citing works
Rozenberg, E., & Freedman, D. (2023) · Machine Learning Science and Technology · doi:10.1088/2632-2153/ace58c · Cited by 2 · see citing works
“By contrast, Fialková et al (2021), Li et al (2021), Thomas et al (2021), Fu et al (2022)use approaches based on reinforcement learning.”
BackgroundREINFORCE-ING Chemical Language Models for Drug Discovery
Thomas, M. C., Bou, A., Gómez-Tamayo, J. C., Tresadern, G., Ahmad, M., & Fabritiis, G. D. (2025) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.5c02053 · Cited by 2 · see citing works
“The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jcim.5c02053.”
InfluentialSelf-citationZang, J., Wang, S., Xu, K., Zhan, X., Tian, Y., Yao, X., & Liu, H. (2026) · Wiley Interdisciplinary Reviews Computational Molecular Science · doi:10.1002/wcms.70083 · Cited by 1 · see citing works
“REINVENT combines a SMILES-based recurrent neural network with reinforcement learning and uses Glide docking scores as structural rewards to guide molecular generation when ligand data are limited [204].”
Glycosylated Antibiotics: New Promising Bacterial Efflux Pumps Inhibitors
Kamaz, Z. A., Umar, H. I., Doshi, P., & Suri, P. (2024) · Microbiology Research Journal International · doi:10.9734/mrji/2024/v34i31435 · Cited by 1 · see citing works
D, K. H. M., Thomas, M. C., Kageyama, M., et al. (2023) · ChemRxiv · doi:10.26434/chemrxiv-2023-lbvgn · Cited by 1 · see citing works
Self-citationThe Role of C omputer‐Aided Drug Design in Drug Discovery
van der Voort, S., Bender, A., & Westerman, B. A. (2024) · Computational Drug Discovery · doi:10.1002/9783527840748.ch9 · Cited by 1 · see citing works
Self-citationExplaining and avoiding failures modes in goal-directed generationPreprint
Langevin, M., Vuilleumier, R., & Bianciotto, M. (2021) · doi:10.33774/chemrxiv-2021-4m6b3
Explaining and avoiding failures modes in goal-directed generationPreprint
Langevin, M., Vuilleumier, R., & Bianciotto, M. (2021) · doi:10.33774/chemrxiv-2021-4m6b3-v2
Explaining and avoiding failures modes in goal-directed generationPreprint
Langevin, M., Vuilleumier, R., & Bianciotto, M. (2021) · ChemRxiv · doi:10.26434/chemrxiv-2021-4m6b3
Comprehensive machine learning boosts structure-based virtual screening for PARP1 inhibitorsPreprint
Caba, K., Tran‐Nguyen, V., Rahman, T., & Ballester, P. J. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.03.15.585277
Antibiotic Drug Development Methods
Jasmine, J., & Mohapatra, S. S. (2024) · Springer eBooks · doi:10.1007/978-981-97-9634-2_12
Structural biology and computer-aided drug design
Singh, P., Hassan, S., Mehdi, M. M., Kumar, V. R., & Biswas, K. (2025) · Elsevier eBooks · doi:10.1016/b978-0-443-27461-9.00008-1
Additional file 4 of Papyrus: a large-scale curated dataset aimed at bioactivity predictions
Béquignon, O. J. M., Bongers, B. J., Jespers, W., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26559776
Caba, K., Tran‐Nguyen, V., Rahman, T., & Ballester, P. J. (2024) · Figshare · doi:10.6084/m9.figshare.26703268
Additional file 2 of Papyrus: a large-scale curated dataset aimed at bioactivity predictions
Béquignon, O. J. M., Bongers, B. J., Jespers, W., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26559770
Additional file 1 of Decrypting orphan GPCR drug discovery via multitask learning
Huang, W.-C., Lin, W.-T., Hung, M.-S., Lee, J.-C., & Tung, C.-W. (2024) · Figshare · doi:10.6084/m9.figshare.26671386
D, K. H. M., Thomas, M. C., Kageyama, M., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26643355
Self-citationD, K. H. M., Thomas, M. C., Kageyama, M., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26643361
Self-citationD, K. H. M., Thomas, M. C., Kageyama, M., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26643358
Self-citationAdditional file 3 of Papyrus: a large-scale curated dataset aimed at bioactivity predictions
Béquignon, O. J. M., Bongers, B. J., Jespers, W., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26559773
Thomas, M., O’Boyle, N. M., Bender, A., & de Graaf, C. (2024) · Figshare · doi:10.6084/m9.figshare.26727727
Self-citationAdditional file 1 of Papyrus: a large-scale curated dataset aimed at bioactivity predictions
Béquignon, O. J. M., Bongers, B. J., Jespers, W., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26559767
D, K. H. M., Thomas, M. C., Kageyama, M., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26643349
Self-citationD, K. H. M., Thomas, M. C., Kageyama, M., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26643346
Self-citationD, K. H. M., Thomas, M. C., Kageyama, M., et al. (2024) · Figshare · doi:10.6084/m9.figshare.26643352
Self-citationGenerative artificial intelligence for de novo drug design
Ali, A. M., Qazi, S., Almuqdadi, H. T. A., Neeraj, Majeed, O. S., Majeed, A. A., Temur, A. A., Murshedi, G. H. M. A., & Waheed, M. S. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-44415-9.00018-1
Novel molecular design via a scaffold-aware transformer with multi-scale attention mechanisms
Park, J., & Yoo, S. (2026) · Journal of Cheminformatics · doi:10.1186/s13321-026-01221-6
“SEDiv is defined as the fraction of representative leader molecules selected from a randomly sampled subset of valid and unique generated molecules, where higher values indicate greater structural variety [42].”
A Multidimensional Taxonomy of Generative Artificial Intelligence Approaches for Drug Discovery
Crozes, T., Sabando, V., Soto, A. J., & Ponzoni, I. (2026) · Wiley Interdisciplinary Reviews Data Mining and Knowledge Discovery · doi:10.1002/widm.70104
“Thomas et al. (2021) evaluated the capability of using molecular docking to guide a deep generative model, called REINVENT (Olivecrona et al. 2017), and the authors contrast the model performance and behavior with respect to a ligand-based scoring function.”
Xu, X., Wang, X. L., Luo, D., Yuan, S., & Lin, X. (2026) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.6c02405
- Authors
- Thomas, M., Smith, R. T., O’Boyle, N. M., de Graaf, C., & Bender, A.
- Affiliations
- University of Cambridge (GB)
- Corresponding author
- Bender, Andreas
- Venue
- Journal of Cheminformatics
- Published
- 2021-05-13
- Volume
- 13
- Issue
- 1
- Pages
- 39-39
- ISSN
- 1758-2946 (ISSN-L)
- PMC
- PMC8117600
- OpenAlex
- W3133989558
- Cited by
- 78
- References
- 105
- Citations by year
- 2026: 6 · 2025: 13 · 2024: 21 · 2023: 17 · 2022: 11 · 2021: 10
- FWCI
- 8.96
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- Journal of Cheminformatics (publishedVersion, cc-by)
- ChemRxiv (acceptedVersion, cc-by-nc-nd)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Apollo (University of Cambridge) (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Figshare (submittedVersion, cc-by-nc-nd)
- Apollo
- Apollo (cc-by)
- Topics
- Physical Sciences › Computer Science › Computational Theory and Mathematics › Computational Drug Discovery Methods
- Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Degradation and Inhibitors
- Keywords
- deep generative models, de novo molecule generation, molecular docking, glides, REINVENT, DRD2, G protein-coupled receptors, protein-ligand interactions, hit generation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
