SpliceBind
An isoform-aware druggability prediction benchmark and graph-neural-network framework for binding pockets, curated over 229 kinase pockets spanning 25 protein families, with a structural-biochemical taxonomy distinguishing when pocket-centric druggability methods succeed (e.g., domain deletions) versus inherently fail (allosteric mechanisms).
Composite
48.5
Experimental validation
N/A — retrospective druggability benchmark on curated kinase pockets
Stages
Target IDHit ID
Modalities
small molecule
Task types
classification
Size
kinase_pockets: 229
protein_families: 25
variants_analyzed: 6
splits: {'train': 0, 'val': 0, 'test': 229}
note: 229 kinase pockets across 25 families; six clinically validated resistance variants spanning five mechanism classes; AUROC 0.703 vs P2Rank 0.634
protein_families: 25
variants_analyzed: 6
splits: {'train': 0, 'val': 0, 'test': 229}
note: 229 kinase pockets across 25 families; six clinically validated resistance variants spanning five mechanism classes; AUROC 0.703 vs P2Rank 0.634
License
Other — see arXiv 2606.04020
First release
2026-06-01
Last updated
2026-06-29
Official site
Leaderboard
→ leaderboard
Dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
SpliceBind: Isoform-Aware Prediction of Binding Pocket Druggability · Bryan Cheng, Austin Jin, Joshua Chang · 2026 · paper · doi:N/A — arXiv preprint 2606.04020 · 0 citations
Flags
self_referential
Experts
Groups
—
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
3
coverage
2
maintenance
2
adoption
1
quality
3
accessibility
3
industry_relevance
3
Notes
Introduces an isoform-aware druggability evaluation tied to splice-mediated kinase-inhibitor resistance, a clinically meaningful gap. Honest about hard limits: allosteric resistance mechanisms remain invisible to any pocket-centric method, a boundary the authors explicitly mark. Benchmark is co-released with the authors' own SpliceBind model that leads it, so flagged self_referential and adoption down-weighted. Coverage narrow (229 kinase pockets); modest statistical edge over P2Rank (AUROC 0.703 vs 0.634, p=0.026).