ESM-scan-A tool to guide amino acid substitutions

Area of Science:

• Computational Biology
• Protein Engineering
• Machine Learning in Biochemistry

Background:

• Recent advances in machine learning have revolutionized protein structure prediction and design.
• Predicting site-specific mutational effects on protein stability and function remains a significant computational challenge.
• Accurate free energy calculations for large biological systems and capturing subtle structural changes are difficult.

Purpose of the Study:

• To introduce and validate ESM-Scan, a novel tool for in silico deep mutational scanning.
• To leverage the ESM zero-shot predictor for identifying preferential amino acid substitutions.
• To assess the predictive accuracy of ESM-Scan for protein stability and functionality.

Main Methods:

• Implementation of ESM-Scan utilizing the ESM zero-shot predictor.
• Scanning entire protein sequences for potential amino acid changes.
• Benchmarking ESM-Scan using three public datasets for stability and functionality predictions.
• Experimental validation on a blue-light-activated diguanylate cyclase (MsLadC).

Main Results:

• ESM-Scan accurately predicted the impact of amino acid substitutions on protein stability and function across benchmark datasets.
• Experimental testing on MsLadC confirmed ESM-Scan's ability to identify critical residues, including one involved in allosteric product inhibition.
• The ESM zero-shot model demonstrated strong correlation between predicted fitness and experimental outcomes for amino acid substitutions.

Conclusions:

• ESM-Scan provides a powerful new method for in silico deep mutational scanning, complementing experimental approaches.
• The tool accurately infers the effects of amino acid substitutions, aiding in the understanding of protein function and stability.
• ESM-Scan facilitates rapid, large-scale prediction of mutational effects, accelerating protein engineering and biological research.