henhaobio
Sign in

bioRxivBioinformaticsPreprint

GO-Term Enrichment of Proteome-Scale Docking Profiles as a Biological Search-Space Reduction Layer for Protein Target Discovery

Chase HarmsiD, Judith Klein‐SeetharamaniD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

Identifying protein targets from phenotype-first or mechanism-uncertain compounds remains difficult because proteome-scale docking can generate thousands of structurally plausible interactions per compound. We developed a workflow that converts ranked proteome-scale docking profiles into stable Gene Ontology (GO) Biological Process enrichment signatures and evaluates whether those signatures can reduce the candidate target space while preferentially retaining known drug-target relationships. Docking targets were retained at the PDB-chain level, mapped to unique human gene identities, and analyzed with PANTHER overrepresentation against the screened structural gene universe. GO enrichment was evaluated from the top 25 through 505 ranked proteins in increments of 10, and a stable compound-level GO profile was selected using a Jaccard stability threshold of 0.80 across three consecutive transitions. Benchmarking used Yamanishi drug-target interactions, with 682 mapped compounds assigned to a prespecified development/held-out split (552/130) and evaluated at canonical, mechanistic, and fine-mechanistic biological resolutions. In the full corrected benchmark, 642 compounds with usable canonical GO profiles showed greater within-class than between-class similarity (0.1019 versus 0.0931; delta = 0.0088; 100,000-permutation p = 0.00222), and canonical class explained 1.34% of multivariate GO-profile variation by PERMANOVA (p < 1e-4). Fine-mechanistic labels showed stronger organization in the full dataset (delta = 0.0245; PERMANOVA R-squared = 0.0976; both p < 1e-4). Held-out validation was more modest and metric-dependent: mechanistic labels were significant by PERMANOVA (R-squared = 0.0626, p = 0.0437), whereas the frozen fine-mechanistic analysis showed greater within-class similarity (0.1350 versus 0.1110; p = 0.038) and significant nearest-neighbor recovery (p = 0.0495), but not significant PERMANOVA (p = 0.119). The principal held-out search-space experiment evaluated 107 compounds, 753,492 candidate protein rows, and 424 represented gold-standard targets. A direct GO gate retained 1.26% of candidates while retaining 11.32% of known targets (8.96-fold enrichment); ontology-propagated GO associations retained 5.10% of candidates and 21.46% of known targets (4.21-fold enrichment). At matched candidate-space sizes, GO-associated prioritization retained 27.59% versus 22.41% of known targets at approximately 5% of candidates, 39.39% versus 36.08% at 10%, and 58.73% versus 56.13% at 20%. By contrast, additive protein-level GO reranking was heterogeneous: among 605 evaluable compounds, 21.7% improved their best known-target rank, but mean reciprocal rank decreased from 0.0276 to 0.0189. These results support GO enrichment as an intermediate biological search-space reduction and prioritization layer rather than a universal direct target-scoring function. Keywords: proteome-scale docking; Gene Ontology; target discovery; targetome; PANTHER; target fishing; biological filtering; search-space reduction; Yamanishi benchmark

In-depth summary

Claude Opus 5.5 · xhigh effort · 7 Nov 2026 · from the full text of version 1 (22 Sept 2026) · CC BY 4.0

GO enrichment of proteome-scale docking profiles shrinks the candidate target list: a held-out GO gate kept 11.32% of known targets in 1.26% of candidates; additive reranking was mixed.

Proteome-scale docking yields thousands of predicted interactions per compound, too many to follow up experimentally. The paper adds a workflow that uses the screened structural gene set as the enrichment background and picks a stable enrichment depth, and it tests GO as a search-space reduction layer rather than a per-protein score.

Key results

  1. ShownStable GO Biological Process profiles from docking carry a detectable but weak class signal, and it grows at finer mechanistic resolution in the full dataset (fine-resolution analyses are exploratory).

    Canonical (642 compounds): within-class Jaccard 0.1019 vs 0.0931 between (p = 0.00222), PERMANOVA R-squared 0.0134. Fine-mechanistic (522 compounds, 25 groups): 0.1257 vs 0.1012, PERMANOVA R-squared 0.0976 (p < 1e-4). · 3.2 Biological information was detectable in GO…; Abstract

  2. SuggestedHeld-out replication of class structure is partial and depends on the metric. Broad canonical classes were not recovered with strong evidence.

    Canonical: within vs between 0.1006 vs 0.0943 (p = 0.170), PERMANOVA p = 0.0585. Mechanistic: PERMANOVA R-squared 0.0626 (p = 0.0437). Frozen fine (58 compounds, 15 groups): 0.1350 vs 0.1110 (p = 0.038), nearest-neighbor p = 0.0495, PERMANOVA p = 0.119. · 3.3 Held-out validation supported partial…; Abstract

  3. ShownIn held-out compounds, GO gating retained a small candidate fraction that was strongly enriched for known targets.

    107 compounds, 753,492 candidate rows, 424 represented known targets. Direct GO gate: 1.26% of candidates, 11.32% of targets (8.96-fold). With ontology propagation: 5.10% of candidates, 21.46% of targets (4.21-fold). · 3.5 GO prioritization improved held-out…; Abstract

  4. SuggestedAt matched candidate-space sizes, GO-associated-first prioritization beat random filtering and modestly beat docking rank alone at stringent cutoffs. The two converged at larger fractions.

    Targets retained (GO vs docking vs random): about 5% of candidates, 27.59% vs 22.41% vs 5.01%; 10%, 39.39% vs 36.08% vs 10.01%; 20%, 58.73% vs 56.13% vs 20.01%. Both reached 87.50% at 50%. No significance test is reported for the GO vs docking gap. · 3.5 GO prioritization improved held-out…

  5. ShownAdding a fixed GO overlap bonus to docking rank was heterogeneous and worsened mean reciprocal rank, so the authors do not recommend it as a general target score. It helped some mechanism families, such as monoamine metabolism.

    605 evaluable compounds: median best-known-target rank 436 to 409; 21.7% improved, 73.9% worsened; MRR 0.0276 to 0.0189. Monoamine metabolism (20 compounds): median rank 73.5 to 12, 55% improved. · 3.4 Direct additive GO reranking was…; Abstract

Methods, data and reproducibility

Compounds were docked with AutoDock Vina against 7,518 human PDB chains and mapped to unique genes. PANTHER GO Biological Process enrichment was run on top-ranked slices of 25 to 505 proteins, and a stable profile was chosen by Jaccard stability. Profiles were tested against Yamanishi drug-target labels on a prespecified development/held-out split, and GO was used for candidate gating and reranking.

  • Human PDB structural target library (resolution under 2.5 angstroms, 50% identity grouping), kept at chain level · human · n = 7,518 structures
  • Yamanishi enzyme, GPCR, ion-channel and nuclear-receptor drug-target interaction datasets · n = 5,127 interaction records; 791 drug identifiers; 989 target identifiers
  • Docked benchmark compounds mapped to Yamanishi drugs, split into development and held-out sets · n = 682 compounds (552 development, 130 held-out)
  • Held-out search-space reduction experiment · n = 107 compounds; 753,492 candidate protein rows; 424 represented known targets
Methods & tools
AutoDock Vina, PANTHER overrepresentation test, Gene Ontology Biological Process, Protein Data Bank, UniProt, Yamanishi drug-target datasets, Jaccard similarity, PERMANOVA, Benjamini-Hochberg FDR
Reproducibility gaps
No code or data links are given. The Vina version, docking box, exhaustiveness and pose count are left as an author placeholder in the Methods.
Limitations (5)
  • Docking scores only approximate binding. The structural target set is biased toward proteins with solved structures, and using the screened set as the enrichment background reduces but does not remove this bias. · stated by the authors
  • Held-out fine-mechanistic groups are small (typically 2 to 7 compounds), which limits power and makes the fine held-out PERMANOVA R-squared unstable. A PERMDISP robustness check is still to be run. · stated by the authors
  • The Yamanishi benchmark is not a complete record of relevant targets, and there is no second independent benchmark or prospective biochemical validation. · stated by the authors
  • Docking parameters are not reported (the Methods contain an unfinished placeholder), which makes the results hard to reproduce or judge. · noted in this summary
  • The matched-size gain of GO prioritization over docking alone is small (for example 27.59% vs 22.41%) on 107 held-out compounds. No uncertainty interval or test is given for this comparison. · noted in this summary

Read this if you triage proteome-scale docking output for phenotype-first compounds and want evidence on whether GO enrichment helps narrow candidates without a direct target score.

Citations
1

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISBarsainyan '26Lin '17Pabon '18Wang '20Halder '24Stevenson '23
66% semantic matchBioinformatics
Aryan Amit Barsainyan, Rishikesh Panda, Jose Siguenza+2 more·bioRxiv·Apr 16, 2026