Predicting the effects of mutations on protein solubility using graph convolution network and protein language model...

Area of Science:

• Computational biology
• Protein engineering
• Bioinformatics

Background:

• Protein solubility is crucial and affected by mutations, potentially causing diseases.
• Experimental solubility determination is costly and time-consuming.
• Current in-silico methods rely on evolutionary data and often neglect 3D structure, limiting performance.

Purpose of the Study:

• To develop an efficient and accurate in-silico method for predicting protein solubility changes due to mutations.
• To leverage predicted protein structures and advanced machine learning for improved prediction accuracy.

Main Methods:

• Proposed DeepMutSol, a sequence-based method using Graph Convolutional Networks (GCN).
• Initiated protein graphs using predicted structures from AlphaFold2.
• Employed protein language embeddings for residue representation.
• Pretrained the model on absolute protein solubility to address limited mutation data.

Main Results:

• DeepMutSol demonstrated superior performance compared to state-of-the-art methods in benchmark tests.
• The method successfully differentiated pathogenic mutations in clinically relevant genes from the ClinVar database.
• Predicted solubility changes correlated with mutation pathogenicity.

Conclusions:

• DeepMutSol offers a powerful and accurate tool for predicting mutation-induced protein solubility changes.
• The approach integrates predicted structures and deep learning for enhanced bioinformatics predictions.
• This method has potential applications in disease gene variant interpretation and protein design.