What question did this study set out to answer?

To validate the Luoshu convex-hull framework for predicting protein druggability using 85 DUD-E targets.

May 9, 2026Open Access

Luoshu Convex-Hull Geometry Predicts Protein Druggability with AUC = 0.846 on 85 DUD-E Targets

Key Points

To validate the Luoshu convex-hull framework for predicting protein druggability using 85 DUD-E targets.
Full 85-protein DUD-E benchmark validation
Analysis of geometric features such as sphericity and depth_ratio
Use of C-alpha coordinates and backbone dihedral angles, without machine learning or ligand data.
Composite AUC of 0.846 (95% CI: 0.748-0.929, p<0.001) surpassing SiteMap Dscore by +0.106.
Sphericity AUC = 0.814 and depth_ratio AUC = 0.812 identified as strong predictors.
Druggable pockets are more sphere-like geometrically, as opposed to flat undruggable surfaces.

Abstract

Full 85-protein DUD-E benchmark validation of the Luoshu convex-hull druggability framework. Key results: sphericity+areaᵥol+compact composite AUC=0. 846 (95%CI: 0. 748-0. 929, permutation p<0. 001), exceeding SiteMap Dscore (~0. 74) by +0. 106. Strongest single features: sphericity AUC=0. 814, depthᵣatio AUC=0. 812, chcₘax AUC=0. 781. New finding: pocket sphericity (how sphere-like the binding cavity is) is as informative as pocket depth for druggability prediction. Druggable pockets are geometrically sphere-like; undruggable PPI surfaces are flat. Method uses only C-alpha coordinates and backbone dihedral angles, O (n²) time, no ML/MD/ligand data. Transparent reporting: v1 24-protein proof-of-concept AUC=0. 957 also reported for comparison. Reference code (druggabilitydude. py) included. Extends doi: 10. 5281/zenodo. 20073548

Ask AI

Helpful

Bookmark

View Full Paper