StabLyzeGraph: High-throughput screening of combinatorial mutations using graph neural networks.

Area of Science:

• Protein Engineering
• Computational Biology
• Biotechnology

Background:

• Protein stability is crucial for biotechnology and medicine, but predicting beneficial mutations is challenging.
• Existing methods struggle with vast mutational spaces and integrating diverse data.
• Rational protein design requires efficient tools to identify stabilizing mutations.

Purpose of the Study:

• To develop a novel computational framework, StabLyzeGraph, for protein mutational analysis and classification of stabilizing mutations.
• To leverage Graph Neural Networks (GNNs) for enhanced prediction accuracy by integrating multiple data types.
• To provide a user-friendly and scalable tool for accelerating the discovery of improved protein variants.

Main Methods:

• Representing proteins as graphs, integrating physicochemical properties, evolutionary conservation, and structural information.
• Utilizing Graph Neural Networks (GNNs) for predictive modeling of mutation effects.
• Developing Benchmarking and Screening modules for performance evaluation and mutation identification.

Main Results:

• StabLyzeGraph demonstrated strong predictive performance across 23 diverse datasets.
• The framework accurately identifies and ranks impactful single- and multi-site stabilizing mutations.
• The model classifies beneficial mutation combinations based on learned structural impact.

Conclusions:

• StabLyzeGraph offers a robust and versatile approach to accelerate the discovery of stabilizing mutations.
• The tool enhances rational protein design for creating novel biologics with superior performance.
• This open-source framework is freely available to advance protein engineering research.