af2rave: protein ensemble generation with physics-based sampling.

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• Protein structures are dynamic and exist as ensembles of functionally relevant conformations.
• Experimental resolution of these conformational ensembles is challenging.
• Deep learning models like AlphaFold2 predict ensembles but lack explicit physical validation.

Purpose of the Study:

• To introduce an open-source Python package for automated prediction of alternative protein conformations.
• To integrate machine learning-based structure prediction with physics-driven sampling.
• To efficiently explore local conformational space and identify functional states with minimal prior knowledge.

Main Methods:

• Combines reduced multiple sequence alignment (MSA) AlphaFold2 predictions with biased or unbiased molecular dynamics (MD) simulations.
• Features an automated feature selection module to identify important collective variables.
• Validated on systems including E. coli adenosine kinase (ADK) and human DDR1 kinase.

Main Results:

• Successfully identified distinct functional states for ADK and DDR1 kinase with minimal prior biological knowledge.
• Demonstrated conformational sampling efficiency comparable to long unbiased MD simulations for SARS-CoV-2 spike protein receptor-binding domain.
• Significantly reduced computational cost compared to traditional methods.

Conclusions:

• The package provides a streamlined workflow for generating and analyzing alternative protein conformations.
• Offers an accessible tool for drug discovery and structural biology.
• Enhances the exploration of protein conformational diversity by integrating AI with physical simulations.