Benchmarking and analysis of protein docking performance in Rosetta v3.2

Area of Science:

• Computational Biology
• Structural Biology
• Biochemistry

Background:

• RosettaDock is a key tool for predicting protein-protein interfaces and aiding in protein design.
• Accurate prediction of these interfaces is crucial for understanding biological processes and developing therapeutics.

Purpose of the Study:

• To benchmark the performance of RosettaDock v3.2 against the previous version (v2.3) using the comprehensive Docking Benchmark 3.0.
• To evaluate RosettaDock v3.2's capabilities across diverse protein complex types and varying docking difficulties.
• To identify failure modes and explore new functionalities of RosettaDock v3.2.

Main Methods:

• Benchmarking RosettaDock v2.3 and v3.2 on 116 targets from Docking Benchmark 3.0.
• Classifying targets by complex type (antibody-antigen, enzyme-inhibitor, other) and docking difficulty (rigid-body, medium, difficult).
• Performing local docking perturbations and analyzing failure causes, including binding-induced conformational changes.

Main Results:

• RosettaDock v3.2 achieved more docking funnels (56) than v2.3 (49) across the benchmark.
• v3.2 showed improved success rates for antibody-antigen (63%) and enzyme-inhibitor (62%) complexes compared to 'other' complexes (35%).
• Performance varied by difficulty, with higher success in rigid-body targets (58%) versus medium (30%) and difficult (14%) targets. Binding-induced backbone changes were a major cause of failure.

Conclusions:

• RosettaDock v3.2 offers comparable performance to v2.3 with a slight improvement in docking funnel generation.
• Binding-induced conformational changes represent a significant challenge for current protein docking methods.
• RosettaDock v3.2 demonstrates expanded functionality, including the potential for incorporating small molecules and co-factors, setting a baseline for future interface modeling research.