PoseX: AI Defeats Physics Approaches on Protein-Ligand Cross Docking

TL;DR

PoseX presents an open-source benchmark for protein-ligand docking that jointly evaluates self-docking and cross-docking with a new, training-data–blind dataset (PoseX-SD and PoseX-CD) and a broad set of 23 methods spanning physics-based, AI docking, and AI co-folding approaches. A key contribution is the relaxation post-processing module, which, when integrated with AI methods, yields state-of-the-art pose plausibility and fewer intra- and intermolecular clashes, while pocket information substantially boosts cross-docking performance. The study also highlights persistent chirality issues in several AI co-folding models, mitigated by physics-inspired techniques such as Boltz-1x, and demonstrates that pocket similarity significantly governs generalization, especially in cross-docking. The online leaderboard and public dataset aim to standardize and accelerate progress toward practically reliable docking pipelines for drug discovery.

Abstract

Existing protein-ligand docking studies typically focus on the self-docking scenario, which is less practical in real applications. Moreover, some studies involve heavy frameworks requiring extensive training, posing challenges for convenient and efficient assessment of docking methods. To fill these gaps, we design PoseX, an open-source benchmark to evaluate both self-docking and cross-docking, enabling a practical and comprehensive assessment of algorithmic advances. Specifically, we curated a novel dataset comprising 718 entries for self-docking and 1,312 entries for cross-docking; second, we incorporated 23 docking methods in three methodological categories, including physics-based methods (e.g., Schrödinger Glide), AI docking methods (e.g., DiffDock) and AI co-folding methods (e.g., AlphaFold3); third, we developed a relaxation method for post-processing to minimize conformational energy and refine binding poses; fourth, we built a leaderboard to rank submitted models in real-time. We derived some key insights and conclusions from extensive experiments: (1) AI approaches have consistently outperformed physics-based methods in overall docking success rate. (2) Most intra- and intermolecular clashes of AI approaches can be greatly alleviated with relaxation, which means combining AI modeling with physics-based post-processing could achieve excellent performance. (3) AI co-folding methods exhibit ligand chirality issues, except for Boltz-1x, which introduced physics-inspired potentials to fix hallucinations, suggesting modeling on stereochemistry improves the structural plausibility markedly. (4) Specifying binding pockets significantly promotes docking performance, indicating that pocket information can be leveraged adequately, particularly for AI co-folding methods, in future modeling efforts. The code, dataset, and leaderboard are released at https://github.com/CataAI/PoseX.

Figures (14)

• Figure 1: Self-docking vs. Cross-docking.
• Figure 2: Performance on PoseX-SD and PoseX-CD.
• Figure 3: Cross-docking performance difference on "similar" and "dissimilar" binding pockets.
• Figure S1: (a) The distribution of structures per target shows that every protein adopts at least two distinct conformations, and about half of the targets are represented by just two. (b) The distribution of pocket similarities.
• Figure S2: Performance on Astex.