How Feasible Is Docking of PROTACs to POI-E3L Complexes? Testing Physics-Based and ML-Based Docking Tools

Targeted protein degradation (TPD) is an innovative drug discovery approach that leverages small molecules to induce proximity between a protein of interest (POI) and an E3 ubiquitin ligase (E3L) for selective degradation. Among TPD modalities, proteolysis-targeting chimeras (PROTACs) present unique challenges for computational modeling due to their size, flexibility, and complex binding interactions. This study evaluates the performance of physics-based and machine learning (ML)-based docking tools for modeling PROTAC-mediated ternary complexes in self-docking and cross-docking scenarios. A benchmark of 43 POI-PROTAC-E3L experimentally resolved structures was analyzed using GLIDE, MOE, rDock, DiffDock, and GeoDirDock. Docking strategies took into account van der Waals scaling, hydrogen bond constraints, receptor flexibility via normal mode analysis (NMA), and AlphaFold2 structures. Results show rDock with high sampling outperforming other physics-based tools, while ML-based tools achieved competitive root-mean-square deviation values but gave good results for systems close to those in their training sets and often generated unrealistic poses, emphasizing the need for postprocessing refinement. Receptor flexibility introduced through NMA significantly enhanced docking results. This work establishes performance benchmarks for PROTAC docking and offers guidelines for leveraging docking tools in PROTAC discovery workflows.