Luoshu Pocket Geometry Predicts Protein Druggability: Pocket Shape and Depth as Orthogonal Geometric Signals (v3, n=85)

Key Points

• This study aims to assess how pocket shape and depth can predict the druggability of proteins.
• Evaluated 85 DUD-E proteins using two geometric signals from the Luoshu framework.
• Measured pocket sphericity and depth to generate a composite score for predicting druggability.
• Compared results with existing methods like SiteMap and pseudo-fpocket.
• Achieved an area under the curve (AUC) of 0.851 for the composite score, outperforming SiteMap by +0.111.
• Pocket sphericity alone has an AUC of 0.814; pocket depth contributes to the prediction with an AUC of 0.781.
• Ablation studies confirm sphericity is the dominant predictor for druggability.