Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

Scaling Structure Aware Virtual Screening to Billions of Molecules with SPRINT

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read SPRINT's central claim is that structure-aware protein-language co-embeddings with learned attention pooling make cosine similarity a reliable, ultra-fast predictor of drug-target binding, enabling billion-molecule virtual screens.

desk verdict SPRINT's billion-scale retrieval is real and worth publishing, but the headline LIT-PCBA zero-shot claim doesn't rule out ligand-side leakage from a PubChem-heavy training set. read the letter →

arxiv 2411.15418 v2 pith:5EIPSI6O submitted 2024-11-23 q-bio.BM cs.LG

Andrew T. McNutt , Abhinav K. Adduri , Caleb N. Ellington , Monica T. Dayao , Eric P. Xing , Hosein Mohimani , David R. Koes This is my paper · ORCID
classification q-bio.BMcs.LG
keywords virtualscreeningdrug-targetinteractionpredictionco-embeddingproteinlanguagemodelsstructure-awareembeddingsmulti-headattentionpoolingvectorsearchLIT-PCBA
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that virtual screening no longer needs to be bottlenecked by molecular docking. It presents SPRINT, a model that learns a shared vector space for drugs and protein targets, where binding likelihood is read off as a sigmoid-scaled cosine similarity. The authors claim that by pooling per-residue embeddings from a structure-aware protein language model with learned multi-head attention, SPRINT beats the leading contrastive and docking baselines on the LIT-PCBA benchmark, matches or exceeds prior methods on DTI classification and binding-affinity prediction, and can query a 6.7-billion-molecule library against a whole proteome in minutes. If these results hold, the practical payoff is that screening for off-target effects, repurposing, and new mechanisms of action becomes a fast retrieval task rather than a weeks-long docking campaign.

What carries the argument

The load-bearing object is the drug–target co-embedding space defined by $P(Y=1|Z_d,Z_t)=\sigma(\alpha\, \text{cosine}(Z_d,Z_t))$ with $\alpha=5$. A frozen Morgan fingerprint encoder and a frozen structure-aware protein language model (SaProt) feed modality-specific MLPs whose outputs $Z_d$ and $Z_t$ are the co-embeddings; the protein side uses multi-head attention pooling over per-residue embeddings instead of the average pooling used by ConPLex. Structure is injected through Foldseek tokens computed from AlphaFold2 structures, so the model sees sequence and geometry without explicit docking. The same space serves three tasks: binary interaction classification under cross-entropy, affinity regression when the sigmoid and cosine are replaced by a dot product, and ultra-fast retrieval when the embeddings are indexed in a vector store.

What would settle it

Compute exact and Tanimoto-similarity overlap between the MERGED training molecules and the LIT-PCBA ligands, retrain SPRINT after excluding any training pair whose drug resembles a benchmark ligand, and re-measure AUROC and enrichment factor at 0.5%; a substantial drop would indicate that the reported enrichment came from memorization rather than generalization.

Watch

Extended reading notes

Core claim

On the paper's own terms, SPRINT's discovery is that structure-aware protein language embeddings, when aggregated by a learnable attention mechanism, form a co-embedding space in which cosine distance is a sufficient statistic for drug-target interaction. Trained with binary cross-entropy on the large MERGED dataset containing PubChem, BindingDB, and ChEMBL interactions, the 16-million-parameter model reaches 73.4% AUROC and a 15.90 enrichment factor at a 0.5% false-positive rate on LIT-PCBA in a zero-shot setting, outperforming DrugCLIP, Gnina, and docking baselines. The same co-embedding space doubles as a binding-affinity predictor, matching the top TDC BindingDB Patent leaderboard ensemble, and as a vector-search index that screens 6.7 billion Enamine REAL molecules against the human proteome in 16 minutes. The paper also reports that SPRINT's selected compounds score higher in a docking-based CACHE2 screen than the first random batch of DeepDocking, and that its molecule embeddings improve antibacterial and toxicity property prediction when concatenated with Morgan fingerprints.

Load-bearing premise

The load-bearing assumption is that the LIT-PCBA zero-shot evaluation is free of ligand leakage: only proteins with at least 90% sequence homology were removed from pretraining, and no check is described for whether the same or closely similar drug molecules appear in the PubChem, BindingDB, or ChEMBL training data.

Editorial extensions

If this is right

  • If the reported LIT-PCBA numbers are right, ligand enrichment no longer requires docking: a frozen protein-language representation plus learned pooling outperforms structure-based screeners on a benchmark designed to avoid DUD-E-style bias.
  • Whole-proteome and pan-species screens become cheap enough to run routinely, making off-target prediction and drug repurposing practical at the scale of billions of molecules.
  • The same co-embedding space can be used directly for binding-affinity prediction, so classification, regression, and retrieval need not be separate models.
  • DTI pre-training produces molecule embeddings that add value on top of Morgan fingerprints for antibacterial and toxicity prediction, suggesting the co-embedding space captures target-neighborhood information.
  • SPRINT can replace the initial random sample in iterative docking pipelines, yielding higher-scoring candidates with roughly one-sixth of the docking effort.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the 16-minute screen is reported as query time; a full deployment would also need to account for index construction, memory, and update costs as a 6.7-billion-embedding store grows, which the paper does not quantify.
  • Editorial inference: the zero-shot claim would be much stronger with a ligand-side leakage check; measuring Tanimoto similarity between MERGED training drugs and LIT-PCBA ligands would show whether enrichment comes from generalization or from memorization of related molecules.
  • Editorial inference: because the best-screening attention maps are the least interpretable, the architecture likely separates two useful properties; a future design could train an interpretable pooler for explanation while keeping a high-recall pooler for screening.
  • Editorial inference: the co-embedding recipe may transfer to non-protein targets such as RNA or modified peptides, but SaProt's amino-acid vocabulary and structure tokens would need to be replaced or extended.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The manuscript introduces SPRINT, a deep co-embedding model for drug–target interaction prediction in which Morgan fingerprints and structure-aware protein language model (SaProt) embeddings are projected into a shared space, with interaction probability given by a sigmoid of a scaled cosine similarity and protein residues aggregated by multi-head attention pooling. The authors report state-of-the-art results on the LIT-PCBA virtual screening benchmark, on several DTI classification benchmarks (particularly with the smaller SPRINT-sm variant), and match the top leaderboard method on the TDC BindingDB Patent affinity prediction task. They also demonstrate a large-scale screen of the Enamine REAL database against the human proteome in minutes, analyze attention maps for interpretability, and release code and data.

Significance. If the LIT-PCBA results were leakage-free and the CACHE2 proxy were a valid measure of hit finding, SPRINT would be a significant advance: it combines an interpretable, structure-aware protein representation with a highly scalable vector-retrieval screening pipeline, and the authors provide a reproducible codebase. Strengths include the transparent architecture, the use of shared splits with reported variance in Table 1, the application of a structure-aware PLM in this co-embedding setting, and the concrete large-scale screening demonstration. However, the central virtual-screening claim depends on the LIT-PCBA evaluation being uncontaminated and on a model-selection protocol that does not use the test benchmark; these conditions are not currently established.

major comments (5)
  1. [Section 2.1 and Section 3.1] The zero-shot LIT-PCBA evaluation decontaminates only the protein side. Section 2.1 states that all protein sequences with at least 90% homology to LIT-PCBA were removed from MERGED using MMSeqs2, but no ligand-side filter is described. Since LIT-PCBA activity labels derive from PubChem bioassays and MERGED is 98.31% PubChem (Section 3.1), the same active molecules or close analogs can occur in the training set. Because SPRINT's drug encoder is a Morgan-fingerprint MLP, the model can memorize ligand-label associations and inflate EF and BEDROC regardless of target-specific binding. The authors should report the overlap between LIT-PCBA ligands and MERGED training molecules (at identity and at typical analog thresholds) and re-run the evaluation with these molecules removed, or otherwise demonstrate that the reported metrics are unchanged.
  2. [Section 2.1 and Table C2] The choice of the 3:1 negative sampling ratio is made using LIT-PCBA performance. Table C2 presents LIT-PCBA results for 1:1 and 3:1 sampling, and the 3:1 configuration is then reported as the final model in Table 2. This is a form of model selection on the test benchmark, which invalidates the 'zero-shot' characterization and may inflate the reported numbers if the benchmark also drives hyperparameter selection. The authors should either hold out LIT-PCBA entirely for final evaluation, or show that the chosen ratio was selected on a separate validation set without reference to LIT-PCBA.
  3. [Section 2.2 and Figure 1] The CACHE2 comparison uses Gnina CNN VS docking scores as a proxy for hit-finding performance. The claim that SPRINT 'finds almost three times the number of high-scoring molecular scaffolds' is based on a threshold of CNN VS > 6, not on experimentally confirmed hits. The distributional comparison in Figure 1 shows that SPRINT-selected molecules have higher predicted docking scores than DeepDocking's Batch 0, but this does not establish that SPRINT would produce more hits in the prospective CACHE2 setting. The authors should re-frame this section as a docking-score distribution analysis and remove or temper the hit-finding language.
  4. [Table 1] The text states that attention pooling achieves 'SOTA predictive scores for DTIs on most benchmarks,' but the full SPRINT (16M) model is not the best model on BIOSNAP, Unseen Drugs, Unseen Targets, DAVIS, or BindingDB; the smaller SPRINT-sm (10M) model is. The SOTA claim is only consistently supported for SPRINT-sm on those benchmarks and for SPRINT on MERGED. The text should be revised to attribute each result to the appropriate model variant.
  5. [Table 2] Table 2 reports no standard deviations or replicate runs for the LIT-PCBA AUROC, BEDROC, and EF values. Given the reported sensitivity to random seeds in Section 2.3 and Appendix E, the absence of uncertainty estimates makes it difficult to assess whether the improvements over DrugCLIP and other baselines are statistically meaningful. Please provide confidence intervals or standard errors for the main virtual screening results.
minor comments (5)
  1. [References] The reference for the CACHE2 method [33] is incomplete and should be updated to a proper citation.
  2. [Table captions] Please check the 'T able' spacing artifacts in table captions and ensure the final PDF renders table titles correctly.
  3. [Section 2.2] The sentence 'Crystal structures with bound fragments in the RNA-binding site exist, with PDB ID 5RLZ used for virtual screening' could be clarified to state which structures were used for the CACHE2 screen.
  4. [Section 2.3] The statement that 'all but one' ProtBert heads attend less to binding residues should be checked against Figure 2(a); if all four heads show this pattern, the text should be corrected.
  5. [General] The term 'zero-shot' is used in two different senses (the unseen drugs/targets splits in Table 1 and the LIT-PCBA evaluation); consider defining the term explicitly to avoid confusion.

Circularity Check

1 steps flagged · score 4.0 of 10

LIT-PCBA 'zero-shot' results are partially selected on the same benchmark, but the core SPRINT training objective and other results are self-contained.

  1. fitted input called prediction [Section 2.1 and Appendix C (Table C2)]
    "To evaluate the performance of SPRINT models at virtual screening on LIT-PCBA in the zero-shot setting, we pre-trained the deeper SPRINT (16M) model on the MERGED dataset after removing all protein sequences with ≥ 90% sequence homology to the LIT-PCBA set using MMSeqs2. ... We observed that models trained with more negatives than positives (at a 3:1 ratio), achieved better virtual screening performance, but had less interpretable attention patterns (Tables 1, C2)."

    The LIT-PCBA results in Table 2 for the final SPRINT model (AUROC 73.4, BEDROC 12.3, EF 15.90 at 0.5%) are exactly the SPRINT 3:1 row of Table C2, the ablation table that evaluates negative sampling on LIT-PCBA. The '3:1' configuration was therefore selected because it performed best on this same benchmark, while Section 2.1 presents the evaluation as 'zero-shot.' The reported LIT-PCBA figures are not the performance of a pre-fixed, untouched model; they are the best configuration chosen by using LIT-PCBA itself for model selection. This is a hyperparameter fit to the benchmark, not an independent zero-shot prediction, although the underlying model remains trained on separate DTI data.

full rationale

The paper's central derivation is a supervised co-embedding model: drug and target encoders are projected into a shared space and trained with binary cross-entropy on cosine similarity (Equation 1), with affinity prediction obtained by replacing cosine similarity with a dot product. This is a trained model evaluated on external benchmarks, not a first-principles derivation whose output is equivalent to its input by construction. I found no load-bearing self-citation chain or uniqueness theorem imported from the authors' prior work; references to Gnina and the CACHE2 second-place team are baseline/comparison uses, not justifications of SPRINT's correctness. The one concrete circularity-adjacent step is the use of LIT-PCBA to select the 3:1 negative-sampling ratio before presenting LIT-PCBA results as zero-shot, which partially compromises the independence of that particular claim. The potential ligand-side overlap between LIT-PCBA and the PubChem-dominated MERGED training set is a benchmark leakage risk rather than a derivation-level circularity, so it is noted but not scored as circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on three hand-chosen parameters (sigmoid scaling alpha=5, 3:1 negative sampling, homology thresholds) and five domain assumptions about representation and evaluation. No new physical entities are introduced. The paper is honest about most of these choices, but the 3:1 ratio was tuned on the headline benchmark.

free parameters (3)
  • alpha (cosine similarity scaling) = 5
    Chosen by hand to expand cosine similarity into the sigmoid range; used in Eq. 1 and C2. The model's probability calibration and ranking depend on this value.
  • negative sampling ratio = 3:1
    Selected after observing LIT-PCBA performance (Table C2); this hyperparameter was tuned on the benchmark used for the headline enrichment claim.
  • homology filtering thresholds = 90% for LIT-PCBA removal; 70% identity and 80% coverage for MERGED splits
    Hand-chosen thresholds that define the zero-shot split; the LIT-PCBA SOTA claim depends on this filtering being sufficient to prevent leakage.
assumptions (5)
  • domain assumption Cosine similarity between learned drug and protein co-embeddings is a monotone proxy for binding probability (Eq. 1).
    The entire model reduces DTI prediction to a similarity score; if this assumption fails, enrichment on new targets is not guaranteed.
  • domain assumption FoldSeek structure tokens from AlphaFold2 predicted structures, with pLDDT below 70 masked, preserve the binding-site information needed for DTI prediction.
    The structure-aware advantage in Table 2 relies on predicted structures being informative for small-molecule binding; inaccurate predictions would erase the gap over ProtBert.
  • domain assumption The MERGED labels from PubChem, BindingDB, and ChEMBL are trustworthy enough to train a generalizing co-embedding.
    Training labels are noisy and imbalanced (83.7M negatives); the model's ranking quality inherits this noise.
  • ad hoc to paper Removing proteins with at least 90% homology to LIT-PCBA from the training set makes the LIT-PCBA evaluation zero-shot for drugs as well as proteins.
    Only protein-side homology is controlled; ligand overlap with LIT-PCBA actives is not checked, so the zero-shot assumption may be violated.
  • domain assumption Gnina CNN VS score is a valid proxy for hit-finding in the CACHE2 comparison.
    The claimed CACHE2 enhancement in Figure 1 compares SPRINT-selected molecules by docking score, not by experimental binding; if Gnina scores do not reflect true binding, the enhancement claim is unsupported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scaling Structure Aware Virtual Screening to Billions of Molecules with SPRINT." pith.science (2026). https://pith.science/paper/5EIPSI6O

@misc{pith2026241115418,
  author       = {Pith},
  title        = {Pith review of: Scaling Structure Aware Virtual Screening to Billions of Molecules with SPRINT},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5EIPSI6O}},
  note         = {Machine review of arXiv:2411.15418}
}
read the original abstract

Virtual screening of small molecules against protein targets can accelerate drug discovery and development by predicting drug-target interactions (DTIs). However, structure-based methods like molecular docking are too slow to allow for broad proteome-scale screens, limiting their application in screening for off-target effects or new molecular mechanisms. Recently, vector-based methods using protein language models (PLMs) have emerged as a complementary approach that bypasses explicit 3D structure modeling. Here, we develop SPRINT, a vector-based approach for screening entire chemical libraries against whole proteomes for DTIs and novel mechanisms of action. SPRINT improves on prior work by using a self-attention based architecture and structure-aware PLMs to learn drug-target co-embeddings for binder prediction, search, and retrieval. SPRINT achieves SOTA enrichment factors in virtual screening on LIT-PCBA, DTI classification benchmarks, and binding affinity prediction benchmarks, while providing interpretability in the form of residue-level attention maps. In addition to being both accurate and interpretable, SPRINT is ultra-fast: querying the whole human proteome against the ENAMINE Real Database (6.7B drugs) for the 100 most likely binders per protein takes 16 minutes. SPRINT promises to enable virtual screening at an unprecedented scale, opening up new opportunities for in silico drug repurposing and development. SPRINT is available on the web as ColabScreen: https://bit.ly/colab-screen

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Contrastive Geometric Learning Unlocks Unified Structure- and Ligand-Based Drug Design

    cs.LG 2026-01 conditional novelty 7.0 of 10

    A contrastive geometric model jointly trained on protein–ligand complexes and bioactivity data achieves competitive virtual screening, strong target fishing, and ligand-conditioned pocket prediction without predefined...

Reference graph

Works this paper leans on

58 extracted references · 46 canonical work pages · cited by 1 Pith paper

  1. [1]

    & Koes, D

    Sunseri, J. & Koes, D. R. Virtual screening with gnina 1.0. Molecules 26, 7369 (2021)

    work page 2021

  2. [2]

    & Berger, B

    Singh, R., Sledzieski, S., Bryson, B., Cowen, L. & Berger, B. Contrastive learning in protein language space predicts interactions between drugs and protein targets. Proceedings of the National Academy of Sciences 120, e2220778120 (2023). URL https://www.pnas.org/doi/10.1073/pnas.2220778120. Publisher: Proceedings of the National Academy of Sciences

    doi:10.1073/pnas.2220778120 2023

  3. [3]

    Antimicrobial resistance: global report on surveil- lance (World Health Organization, Geneva, 2014)

    World Health Organization. Antimicrobial resistance: global report on surveil- lance (World Health Organization, Geneva, 2014). URL https://iris.who.int/ handle/10665/112642. Section: xxii, 232 p

    work page 2014

  4. [4]

    K., Bayliss, S

    Colclough, A., Corander, J., Sheppard, S. K., Bayliss, S. C. & Vos, M. Patterns of cross-resistance and collateral sensitivity between clinical antibiotics and natural antimicrobials. Evolutionary applications 12, 878–887 (2019)

    work page 2019

  5. [5]

    Morgan, H. L. The Generation of a Unique Machine Description for Chemi- cal Structures-A Technique Developed at Chemical Abstracts Service. Journal of Chemical Documentation 5, 107–113 (1965). URL https://doi.org/10.1021/ c160017a018. Publisher: American Chemical Society

    work page 1965

  6. [6]

    & Linial, M

    Brandes, N., Ofer, D., Peleg, Y., Rappoport, N. & Linial, M. ProteinBERT: a universal deep-learning model of protein sequence and function. Bioinformatics 38, 2102–2110 (2022). URL https://doi.org/10.1093/bioinformatics/btac020

    doi:10.1093/bioinformatics/btac020 2022

  7. [7]

    Rives, A. et al. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences. Proceedings of the National Academy of Sciences 118 (2021). URL http://dx.doi.org/10.1073/pnas. 2016239118

    doi:10.1073/pnas 2021

  8. [8]

    Su, J. et al. SaProt: Protein Language Modeling with Structure-aware Vocabulary (2024). URL https://www.biorxiv.org/content/10.1101/2023.10.01.560349v4. Pages: 2023.10.01.560349 Section: New Results

    work page doi:10.1101/2023.10.01.560349v4 2024

Show all 58 references
  1. [9]

    Gao, B. et al. Drugclip: Contrasive protein-molecule representation learning for virtual screening. Advances in Neural Information Processing Systems 36 (2024)

    2024

  2. [10]

    Jia, Y. et al. Deep contrastive learning enables genome-wide virtual screening. bioRxiv 2024–09 (2024)

    2024

  3. [11]

    Meller, A. et al. Predicting locations of cryptic pockets from single protein struc- tures using the pocketminer graph neural network. Nature Communications 14, 1177 (2023). 23

    2023

  4. [12]

    Van Kempen, M. et al. Fast and accurate protein structure search with foldseek. Nature biotechnology 42, 243–246 (2024)

    2024

  5. [13]

    Chroma - the open-source embedding database

    Chroma. Chroma - the open-source embedding database. https://www.trychroma.com/ (2024)

    2024

  6. [14]

    Zhang, Z. et al. Protein language models learn evolutionary statistics of interacting sequence motifs. bioRxiv 2024–01 (2024)

    2024

  7. [15]

    Touvron, H. et al. Augmenting convolutional networks with attention-based aggregation. arXiv preprint arXiv:2112.13692 (2021)

    2021 arXiv

  8. [16]

    Golts, A. et al. A large dataset curation and benchmark for drug target interaction. arXiv preprint arXiv:2401.17174 (2024)

    2024 arXiv

  9. [17]

    Kim, S. et al. Pubchem 2023 update. Nucleic acids research 51, D1373–D1380 (2023)

    2023

  10. [18]

    Gilson, M. K. et al. Bindingdb in 2015: a public database for medicinal chemistry, computational chemistry and systems pharmacology. Nucleic acids research 44, D1045–D1053 (2016)

    2016

  11. [19]

    Gaulton, A. et al. The chembl database in 2017. Nucleic acids research 45, D945–D954 (2017)

    2017

  12. [20]

    & Jain, A

    Spitzer, R. & Jain, A. N. Surflex-dock: Docking benchmarks and real-world application. Journal of computer-aided molecular design 26, 687–699 (2012)

    2012

  13. [21]

    Halgren, T. A. et al. Glide: a new approach for rapid, accurate docking and scor- ing. 2. enrichment factors in database screening. Journal of medicinal chemistry 47, 1750–1759 (2004)

    2004

  14. [22]

    Zhang, X. et al. Planet: A multi-objective graph neural network model for protein- ligand binding affinity prediction. biorxiv (2023)

    2023

  15. [23]

    McNutt, A. T. et al. Gnina 1.0: molecular docking with deep learning. Journal of cheminformatics 13, 43 (2021)

    2021

  16. [24]

    & Ozkirimli, E

    ¨Ozt¨ urk, H.,¨Ozg¨ ur, A. & Ozkirimli, E. Deepdta: deep drug–target binding affinity prediction. Bioinformatics 34, i821–i829 (2018)

    2018

  17. [25]

    Brocidiacono, M. et al. Bigbind: learning from nonstructural data for structure- based virtual screening. Journal of Chemical Information and Modeling 64, 2488– 2495 (2023)

    2023

  18. [26]

    & Rognan, D

    Tran-Nguyen, V.-K., Jacquemard, C. & Rognan, D. Lit-pcba: an unbiased data set for machine learning and virtual screening. Journal of chemical information and modeling 60, 4263–4273 (2020). 24

    2020

  19. [27]

    & Heifets, A

    Wallach, I. & Heifets, A. Most ligand-based classification benchmarks reward memorization rather than generalization. Journal of chemical information and modeling 58, 916–932 (2018)

    2018

  20. [28]

    & S¨ oding, J

    Steinegger, M. & S¨ oding, J. Mmseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nature biotechnology 35, 1026–1028 (2017)

    2017

  21. [29]

    Huang, K. et al. Artificial intelligence foundation for therapeutic science. Nature chemical biology 18, 1033–1036 (2022)

    2022

  22. [30]

    Lam, H. T. et al. Otter-knowledge: benchmarks of multimodal knowledge graph representation learning from different sources for drug discovery. arXiv preprint arXiv:2306.12802 (2023)

    2023 arXiv

  23. [31]

    Ackloo, S. et al. Cache (critical assessment of computational hit-finding exper- iments): A public–private partnership benchmarking initiative to enable the development of computational methods for hit-finding.Nature Reviews Chemistry 6, 287–295 (2022)

    2022

  24. [32]

    Yazdani, S. et al. Genetic variability of the sars-cov-2 pocketome. Journal of Proteome Research 20, 4212–4215 (2021)

    2021

  25. [33]

    of pittsburgh] computational methods - cache2

    Koes, dunn [u. of pittsburgh] computational methods - cache2. URL https:// cache-challenge.org/challenges/app/63160acad0a52

  26. [34]

    Gentile, F. et al. Deep docking: a deep learning platform for augmentation of structure based drug discovery. ACS central science 6, 939–949 (2020)

    2020

  27. [35]

    & Dunbrack Jr, R

    Modi, V. & Dunbrack Jr, R. L. A structurally-validated multiple sequence alignment of 497 human protein kinase domains. Scientific reports 9, 19790 (2019)

    2019

  28. [36]

    & Joshi, R

    Setiya, A., Jani, V., Sonavane, U. & Joshi, R. Moltoxpred: small molecule toxicity prediction using machine learning approach.RSC advances 14, 4201–4220 (2024)

    2024

  29. [37]

    Terlouw, B. R. et al. Mibig 3.0: a community-driven effort to annotate experimen- tally validated biosynthetic gene clusters. Nucleic acids research 51, D603–D610 (2023)

    2023

  30. [38]

    Landrum, G. et al. Rdkit: Open-source cheminformatics. https://www.rdkit.org (2024)

    2024

  31. [39]

    Jumper, J. et al. Highly accurate protein structure prediction with alphafold. nature 596, 583–589 (2021)

    2021

  32. [40]

    Davis, M. I. et al. Comprehensive analysis of kinase inhibitor selectivity. Nature biotechnology 29, 1046–1051 (2011). 25

    2011

  33. [41]

    Liu, T., Lin, Y., Wen, X., Jorissen, R. N. & Gilson, M. K. Bindingdb: a web-accessible database of experimentally determined protein–ligand binding affinities. Nucleic acids research 35, D198–D201 (2007)

    2007

  34. [42]

    & Leskovec, J

    Zitnik, M., Sosic, R. & Leskovec, J. Biosnap datasets: Stanford biomedical net- work dataset collection. Note: http://snap. stanford. edu/biodata Cited by 5 (2018)

    2018

  35. [43]

    Huang, K., Xiao, C., Glass, L. M. & Sun, J. Moltrans: molecular interaction transformer for drug–target interaction prediction. Bioinformatics 37, 830–836 (2021)

    2021

  36. [44]

    Li, J. et al. Leak proof pdbbind: A reorganized dataset of protein-ligand complexes for more generalizable binding affinity prediction. arXiv preprint arXiv:2308.09639 (2023)

    2023

  37. [45]

    Frolova, D. et al. Mulan: Multimodal protein language model for sequence and structure encoding. bioRxiv 2024–05 (2024)

    2024

  38. [46]

    Wang, D. et al. S-plm: Structure-aware protein language model via contrastive learning between sequence and structure. bioRxiv (2024). URL https://www. biorxiv.org/content/early/2024/01/28/2023.08.06.552203

    2024

  39. [47]

    Zhou, G. et al. Uni-mol: A universal 3d molecular representation learning framework (2023)

    2023

  40. [48]

    Ji, X. et al. Uni-mol2: Exploring molecular pretraining model at scale. arXiv preprint arXiv:2406.14969 (2024)

    2024 arXiv

  41. [49]

    & Zou, J

    Simon, E. & Zou, J. Interplm: Discovering interpretable features in protein language models via sparse autoencoders. bioRxiv 2024–11 (2024)

    2024

  42. [50]

    G., Zotchev, S

    Atanasov, A. G., Zotchev, S. B., Dirsch, V. M. & Supuran, C. T. Natural products in drug discovery: advances and opportunities. Nature reviews Drug discovery 20, 200–216 (2021)

    2021

  43. [51]

    Kishimoto, A. et al. MHG-GNN: Combination of Molecular Hypergraph Gram- mar with Graph Neural Network (2023). URL http://arxiv.org/abs/2309.16374. ArXiv:2309.16374 [cs]

    2023 arXiv

  44. [52]

    Ross, J. et al. Large-Scale Chemical Language Representations Capture Molec- ular Structure and Properties (2022). URL http://arxiv.org/abs/2106.09553. ArXiv:2106.09553 [cs, q-bio]

    2022 arXiv

  45. [53]

    Stokes, J. M. et al. A deep learning approach to antibiotic discovery. Cell 180, 688–702 (2020). 26

    2020

  46. [54]

    & Mohimani, H

    Mongia, M., Guler, M. & Mohimani, H. An interpretable machine learning approach to identify mechanism of action of antibiotics. Scientific Reports 12, 10342 (2022)

    2022

  47. [55]

    & De Fabritiis, G

    Cremer, J., Medrano Sandonas, L., Tkatchenko, A., Clevert, D.-A. & De Fabritiis, G. Equivariant graph neural networks for toxicity prediction. Chemical Research in Toxicology 36, 1561–1573 (2023)

    2023

  48. [56]

    Mirdita, M. et al. Colabfold: making protein folding accessible to all. Nature methods 19, 679–682 (2022)

    2022

  49. [57]

    Liu, Z. et al. Forging the basis for developing protein–ligand interaction scoring functions. Accounts of chemical research 50, 302–309 (2017)

    2017

  50. [58]

    A., Ovchinnikov, S

    Gazizov, A., Lian, A., Goverde, C. A., Ovchinnikov, S. & Polizzi, N. F. Af2bind: Predicting ligand-binding sites using the pair representation of alphafold2. bioRxiv 2023–10 (2023). 27

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.