Shen et al. 2025, Chemical Science (Europe PMC full text)

What this source contributed

Records in this release that cite this source. A record can cite several sources.

• 3 methods
• 55 configurations
• 3 benchmarks
• 7 protocols
• 5 datasets
• 76 evaluations
• 750 results
• 11 evidence claims
• 1 use case

Results

Values as the source printed them. Compare results only under compatible protocols, datasets and metrics.

Claims

Statements this source supports, each with where in the source it was checked.

• Actives-to-decoys ratio and decoy construction differ by orders of magnitude between the screening benchmarks: Reported finding DEKOIS 2.0 holds 40 experimentally validated actives and 1200 property-matched decoys selected from ZINC per target, a ratio of 1:30. GPCRrecent holds BindingDB actives with decoys generated by a conditional recurrent neural network (TocoDecoy) at 1:50. LIT-PCBA's negatives are experimentally measured inactives, and its official per-target counts give ratios from 13 actives to 4,378 inactives (ESR_ago, about 1:337) and 88 to 3,820 (ESR_antago, about 1:43) up to 17 to 311,748 (ADRB2, about 1:18,000). DUD-E pairs each of 22,886 clustered ligands with 50 property-matched decoys. An enrichment factor is a ratio against random selection at a fixed fraction of the list, so it moves with both the ratio and the realism of the negatives; values from these benchmarks are not comparable with one another. (Shen et al. 2025 Methods 'Datasets' and 'Evaluation metrics'; LIT-PCBA official page data content table)
• Hit-discovery use case: evidence_checked names this review: Evidence checked {"date": "2026-10-22", "cutoff_date": "2026-10-22", "records": ["data/omics/lit-search-20261022-uc-hitdisc/report.md", "data/omics/use-case-coverage-hitdisc-20261022/research.md", "data/omics/search-ledger.jsonl", "docs/reviews/use-cases/hitdisc-2026-10-22.md"]} (docs/reviews/use-cases/hitdisc-2026-10-22.md)
• LIT-PCBA page source: totals-conflict concern now names both artifact hashes: Evidence concerns[0]: message The official page (data content table, artifact a2203612b553236d7f01da5573389d439d2f7aa0436080412f27da882e086e5e) gives 7761 actives and 382674 unique inactives for the 15 LIT-PCBA target sets, while Shen et al. 2025 (Methods 'Datasets', artifact 11a4cee97057d9e72f9393024d22fdd169e3134f493367eee2f7869facf9ccf8) state 10030 confirmed actives and 2798737 confirmed inactives for the full set. The two counts are not reconciled by either source; the difference is most likely unique against per-target inactives, but neither source says so. EF and NEF both depend on this ratio, so the conflict is left unresolved and marked rather than reconciled. (LIT-PCBA official page data content table; Shen et al. 2025 Methods 'Datasets')
• Relevance of hitdisc-20261022-protocol-shen-dekois-builtin to use-case-hit-discovery-virtual-screening: Assessed by DEKOIS 2.0 screening with co-folding confidence scores against docking tools (Shen et al. 2025 Table 1) (hitdisc-20261022-protocol-shen-dekois-builtin: see its source_locator)
• Relevance of hitdisc-20261022-protocol-shen-dekois-rescore-af3 to use-case-hit-discovery-virtual-screening: Assessed by DEKOIS 2.0 screening by rescoring AF3-predicted complexes (Shen et al. 2025 Table 2, AF3 block) (hitdisc-20261022-protocol-shen-dekois-rescore-af3: see its source_locator)
• Relevance of hitdisc-20261022-protocol-shen-dekois-rescore-protenix to use-case-hit-discovery-virtual-screening: Assessed by DEKOIS 2.0 screening by rescoring Protenix-predicted complexes (Shen et al. 2025 Table 2, Protenix block) (hitdisc-20261022-protocol-shen-dekois-rescore-protenix: see its source_locator)
• Relevance of hitdisc-20261022-protocol-shen-gpcrrecent-6 to use-case-hit-discovery-virtual-screening: Assessed by GPCRrecent screening (the six targets whose first structures postdate the Boltz-2 training cutoff) (Shen et al. 2025 Table 3) (hitdisc-20261022-protocol-shen-gpcrrecent-6: see its source_locator)
• Relevance of hitdisc-20261022-protocol-shen-gpcrrecent-16 to use-case-hit-discovery-virtual-screening: Assessed by GPCRrecent screening (16 GPCR targets) (Shen et al. 2025 Table 3) (hitdisc-20261022-protocol-shen-gpcrrecent-16: see its source_locator)
• Relevance of hitdisc-20261022-protocol-shen-litpcba5 to use-case-hit-discovery-virtual-screening: Assessed by LIT-PCBA five-target screening (Shen et al. 2025 Table 4) (hitdisc-20261022-protocol-shen-litpcba5: see its source_locator)
• Relevance of hitdisc-20261022-protocol-shen-speed to use-case-hit-discovery-virtual-screening: Assessed by Co-folding time per compound on five DEKOIS 2.0 targets (Shen et al. 2025 Table 5) (hitdisc-20261022-protocol-shen-speed: see its source_locator)
• Summary of the evidence for use-case-hit-discovery-virtual-screening: Summary Three sources in this store score machine-learning and co-folding methods against docking on the same screening data, and what the numbers say depends on how the negatives were built. On DEKOIS 2.0, whose 79 targets each pair 40 validated actives with 1,200 property-matched decoys selected from ZINC (1:30), the co-folding confidence scores enrich strongly and beat the docking arms. AF3 min-iPAE gives a mean EF1% of 23.15 and Boltz-2 min-iPAE 22.73, against 12.01 for the Glide SP row quoted from another paper, 5.16 for Gnina Vina and 2.94 for TankBind; the best rescoring of a co-folded pose, RTMScore on AF3 poses, reaches 21.34. The authors flag their own Boltz-2 affinity arms (mean EF1% 23.54 and 25.84) as likely over-estimates because Boltz-2 drew affinity data from the same public databases that source DEKOIS 2.0. On GPCRrecent, 16 targets whose first crystal structures were released after 2022 with decoys generated by a conditional recurrent neural network at 1:50, every value falls and the margin narrows: Boltz-2 min-iPAE gives a mean EF1% of 13.96 against 5.02 for Glide SP and 3.65 for Gnina Vina. On the five LIT-PCBA targets where the inactives were experimentally measured, the margin largely disappears: averaged over the five targets Boltz-2 min-iPAE gives EF1% 9.30 and NEF1% 0.100 against Glide SP's 6.79 and 0.079, and on TP53 both Boltz-2 scores give EF1% 0.00 while Protenix ipTM gives 5.00 and Glide SP 3.75. A separate audit of LIT-PCBA reports 323 active training-validation analogue pairs at ECFP4 Tanimoto at or above 0.6 in ALDH1 alone and 2,491 2D-identical inactives shared between training and validation across targets, and concludes that a parameter-free memorisation baseline matches or exceeds published deep-learning results there, so every LIT-PCBA value above carries that limit. Where the negatives are compounds a docking screen already ranked highly and every compound was synthesised and assayed, the ordering changes. On the AmpC list (247 inhibitors among 1,293 tested compounds from a 1.7-billion molecule screen) DOCK3.7 separates binders best, AUC 76.5, against 68.1 for Boltz-2 pIC50 and 60.5 for AF3 L-pLDDT. On the dopamine D4 list (205 binders among 541) docking and Boltz-2 are indistinguishable at the precision printed (both about 71) and AF3 L-pLDDT is worse than random at 46.4. Only on the sigma-2 list (201 binders among 506) does a co-folding score lead, Boltz-2 pIC50 at 83.8 against DOCK3.7's 78.8, with AF3 L-pLDDT at 56.1. In the one blinded prospective challenge, CACHE #3 on the SARS-CoV-2 Nsp3 macrodomain, 23 workflows selected 1,739 compounds that were synthesised and assayed; 28 advanced to hit expansion and 29 confirmed hits scored above 15, four of them judged chemically novel. Twelve of the 23 workflows returned no confirmed hit from their own selection of 41 to 99 compounds. When every workflow was applied to the same merged 1,739-compound library, WF1696, which the article describes as a purely physics-based method, captured 2 of the 4 Tier 1 and 9 of the 25 Tier 2 confirmed hits from 99 predicted actives, which the authors report as did better than any other method in retrieving hits from the collection of all molecules selected in Round 1. Cost bounds all of this: co-folding one protein-ligand pair takes 7.79 to 34.58 seconds per compound on one NVIDIA H100 with pre-computed MSAs, which is why the LIT-PCBA arm covers five targets rather than fifteen. No docking tool was timed on the same hardware, no ligand-similarity or pharmacophore baseline is scored anywhere in this batch, and no result here estimates the hit rate of an untested library. (Shen et al. 2025 Tables 1-5; Kim, Correy, Hall et al. 2026 Results; Herasymenko et al. 2026 Results and sheet S8; arXiv:2507.21404v2 Abstract)

Models, benchmarks and other records

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.