RL-MLZerD: Multimeric protein docking using reinforcement learning.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Protein complexes are essential for numerous cellular processes.
• Determining the quaternary structure of protein complexes is crucial for understanding molecular mechanisms.
• Existing computational protein docking methods primarily focus on two-chain interactions, limiting their application to larger complexes.

Purpose of the Study:

• To develop a novel computational method for predicting the quaternary structures of multi-chain protein complexes.
• To address the limitations of existing two-chain docking methods in modeling larger assemblies.

Main Methods:

• Introduction of RL-MLZerD, a method utilizing reinforcement learning (RL) for multi-chain protein complex assembly.
• Modeling the multi-chain assembly process as a series of episodes involving the selection and integration of pairwise docking models within an RL framework.
• Utilizing RL to identify plausible pairwise models that fit well with other subunits in a complex.

Main Results:

• RL-MLZerD demonstrated superior modeling performance on a benchmark dataset of protein complexes with three to five chains compared to other multiple docking methods.
• The method's performance was evaluated using various criteria, showing effectiveness except when compared to AlphaFold-Multimer in unbound docking scenarios.
• The study identified that the docking order of multi-chain complexes can be predicted by analyzing preferred paths within the RL computation.

Conclusions:

• RL-MLZerD offers a promising advancement in computational protein docking for predicting the structures of multi-chain complexes.
• The reinforcement learning approach effectively handles the complexity of assembling multiple protein subunits.
• The method not only predicts structures but also provides insights into the natural docking order of complex assembly.