THPLM: a sequence-based deep learning framework for protein stability changes prediction upon point variations using...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• Accurate prediction of protein thermodynamic stability is crucial for protein and drug design.
• Existing structure-based and sequence-based methods have limitations in representing global sequence effects on stability changes.
• Advancements in protein language models (PLMs) offer new ways to capture structural information from sequences.

Purpose of the Study:

• To develop a novel sequence-based deep learning model for predicting protein stability changes.
• To leverage protein language models (PLMs) for improved representation of protein sequences.
• To assess the performance of the proposed model against existing methods.

Main Methods:

• Utilized Meta's ESM-2, a powerful protein language model, for sequence representation.
• Developed THPLM, a sequence-based deep learning model incorporating a convolutional neural network.
• Evaluated THPLM's performance in predicting protein stability changes.

Main Results:

• THPLM achieved comparable or superior performance to existing sequence-based and structure-based methods.
• The model effectively utilizes PLM-generated sequence representations for stability prediction.
• PLM's representational capacity enhances protein function prediction capabilities.

Conclusions:

• THPLM offers a robust and effective sequence-based approach for predicting protein stability changes.
• The study highlights the potential of PLMs in advancing computational protein design and analysis.
• The developed model and code are publicly available for further research and application.