MPNN Protein Variant Generation for Thermostability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computational methods for generating thermostable protein variants are labor-intensive, rely heavily on trial-and-error, and often disrupt critical conserved regions or active sites, lacking a systematic approach to evaluate dynamic thermal stability.
Innovation Solution
A method using a message-passing neural network (MPNN) to generate mutations in solvent-exposed and loop regions while preserving conserved residues, combined with molecular dynamics simulations to assess thermal stability, ensuring functional integrity and thermostability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If mutations are introduced across the entire protein sequence to explore variants, then the diversity of protein variants increases, but the risk of disrupting conserved regions and active sites increases
Solution Approach 1:
The protein sequence is segmented into mutable regions (solvent-exposed residues and loop regions) and non-mutable regions (conserved residues and active sites). This segmentation allows mutations to be introduced only in specific segments while preserving critical functional regions, thus maintaining protein function while generating diverse variants.
Solution Approach 2:
Different regions of the protein are assigned different mutation tolerances based on their structural and functional properties. Solvent-exposed residues and loop regions are designated as mutation-tolerant, while conserved residues and active sites are protected from mutation. This local differentiation enables targeted mutagenesis that preserves overall protein function.
2Productivity
If static structure prediction is used to evaluate mutant proteins, then the evaluation process is fast and simple, but the accuracy of stability assessment under thermal conditions decreases
Solution Approach 1:
Molecular dynamics simulations are performed in advance to capture the dynamic behavior of mutant proteins under thermal stress conditions before final selection. This preliminary dynamic analysis provides accurate stability assessments and guides the selection of thermostable variants, ensuring both speed and precision in the evaluation process.
Data Source
AI summary
A system and method for generating thermostable variants of a protein is disclosed. The system receives a three-dimensional structure of a target protein and identifies mutable regions, including solvent-exposed residues and loop regions. Conserved and active site residues are excluded from mutation through a fixed-position mask. A message-passing neural network (MPNN) generates mutant sequences at unmasked positions, executed under multiple temperature parameters. Design scores based on Shannon entropy and log probability are computed, and high-confidence variants are selected. Predicted structures for selected variants are evaluated using structural and sequence-based features to compute stability scores. A ranked list of thermostable variants is generated. Top candidates undergo molecular dynamics simulations to compute dynamic metrics such as RMSD, radius of gyration, SASA, and ddG, and are re-ranked accordingly. The system enables accurate, constraint-driven protein design with high structural and functional fidelity, suitable for industrial and therapeutic applications.