PoseX: AI Defeats Physics Approaches on Protein-Ligand Cross Docking (2505.01700v2)

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., Schr\"odinger 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.

• The paper introduces PoseX, a new benchmark showing AI docking methods outperform traditional physics-based approaches in protein-ligand docking accuracy.
• A novel energy minimization technique (relaxation) significantly improved AI-generated poses, enhancing consistency with physicochemical principles.
• While AI methods excel, challenges remain, particularly for AI co-folding methods struggling with ligand chirality, highlighting areas for future research in drug discovery.

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

In the field of computational biochemistry, the paper titled "PoseX: AI Defeats Physics Approaches on Protein-Ligand Cross Docking" presents a detailed examination of protein-ligand docking, a crucial task in drug discovery. This paper introduces PoseX, an open-source benchmark designed to enhance the evaluation of self-docking and cross-docking algorithms more practically and comprehensively.

Methodology

The paper outlines the creation of the PoseX benchmark, featuring a curated dataset with 718 entries for self-docking and 1,312 entries for cross-docking. This dataset is instrumental in comparing the performance of 22 distinct docking methods classified into three categories:

1. Traditional Physics-Based Methods: Examples include Schrödinger Glide and AutoDock Vina, which rely on physics-based scoring functions and sampling algorithms.
2. AI Docking Methods: Techniques such as DiffDock and Uni-Mol employ advanced deep learning (DL) models to predict ligand binding poses.
3. AI Co-Folding Methods: This category, including tools like AlphaFold3, considers the structural changes of both protein and ligand, aiming to simulate their interactions more accurately.

To complement these methods, the research proposes a novel energy minimization technique (relaxation method) for refining AI-generated binding poses, enhancing consistency with physicochemical principles.

Major Findings

The paper reveals several insightful findings:

• AI Docking Superiority: AI methods have surpassed traditional approaches in overall docking accuracy, evidenced by root-mean-square deviation (RMSD) metrics. The paper specifically highlights that AI models, once plagued by generalization issues, have shown marked improvements.
• Post-Processing Impact: The relaxation method proved beneficial, especially for AI docking methods, by alleviating stereochemical deficiencies and achieving high docking accuracy.
• AI Co-Folding Challenges: Despite the improvements, AI co-folding methods encountered persistent ligand chirality issues, which relaxation could not resolve.

Implications

The implications of these findings are profound both practically and theoretically. Practically, they suggest that integrating AI in docking processes can significantly enhance the efficiency and accuracy of drug discovery protocols. Theoretically, these insights highlight the potential for further research in improving chirality handling within AI models, suggesting areas ripe for innovation in biomolecular simulation.

Moreover, the real-time leaderboard released with the PoseX benchmark fosters a dynamic environment where researchers can evaluate their models against standardized criteria, enhancing transparency and collaborative improvement.

Future Directions

Looking ahead, future developments in AI-driven protein-ligand docking could explore enhancing model robustness against novel chemotypes, optimizing ligand flexibility, and implementing joint evaluation frameworks that consider both structural and binding affinity predictions. Such advancements could revolutionize drug design, shortening development cycles and increasing the potential for therapeutic breakthroughs.

In conclusion, this paper provides a comprehensive framework for evaluating protein-ligand docking methods, underscoring the transformative potential of AI in biomedical research. As AI models continue to evolve, they are poised to further integrate into the foundational methodologies of drug discovery, driving both scientific and practical advancements in the field.