The paper "Deep Confident Steps to New Pockets: Strategies for Docking Generalization" addresses the challenge of accurate blind docking, which is crucial for drug discovery and biological research. The authors develop DOCKGEN, a new benchmark based on ligand-binding domains of proteins, to rigorously assess the generalizability of docking methods. They find that existing machine learning (ML)-based docking models have weak generalization abilities and propose several strategies to improve this, including scaling data and model size, integrating synthetic data, and introducing CONFIDENCE BOOTSTRAPPING, a self-training paradigm that leverages the interaction between diffusion and confidence models. CONFIDENCE BOOTSTRAPPING significantly enhances the ability of ML-based docking methods to dock to unseen protein classes, bringing them closer to accurate and generalizable blind docking. The paper also includes a detailed analysis of the scaling laws of ML-based docking and demonstrates the effectiveness of these strategies through experiments on the DOCKGEN benchmark.The paper "Deep Confident Steps to New Pockets: Strategies for Docking Generalization" addresses the challenge of accurate blind docking, which is crucial for drug discovery and biological research. The authors develop DOCKGEN, a new benchmark based on ligand-binding domains of proteins, to rigorously assess the generalizability of docking methods. They find that existing machine learning (ML)-based docking models have weak generalization abilities and propose several strategies to improve this, including scaling data and model size, integrating synthetic data, and introducing CONFIDENCE BOOTSTRAPPING, a self-training paradigm that leverages the interaction between diffusion and confidence models. CONFIDENCE BOOTSTRAPPING significantly enhances the ability of ML-based docking methods to dock to unseen protein classes, bringing them closer to accurate and generalizable blind docking. The paper also includes a detailed analysis of the scaling laws of ML-based docking and demonstrates the effectiveness of these strategies through experiments on the DOCKGEN benchmark.