TemStaPro: protein thermostability prediction using sequence representations from protein language models.

Area of Science:

• Computational Biology
• Biophysics
• Machine Learning

Background:

• Predicting protein thermostability from amino acid sequences is crucial for biological and industrial applications.
• Deep learning and machine learning offer powerful tools for sequence-based predictions.
• Protein language models (pLMs) excel at capturing complex sequence features.

Purpose of the Study:

• To develop a robust method for predicting protein thermostability using sequence data.
• To leverage transfer learning and pLMs for enhanced prediction accuracy.
• To apply the developed method to predict the thermostability of CRISPR-Cas Class II effector proteins (C2EPs).

Main Methods:

• Applied transfer learning using embeddings from large, pre-trained protein language models (pLMs).
• Trained and validated the prediction model on over one million protein sequences with annotated growth temperatures.
• Developed the TemStaPro (Temperatures of Stability for Proteins) software for prediction.

Main Results:

• Achieved high-performing protein thermostability predictions.
• Demonstrated significant differences in thermostability among C2EP groups.
• Validated predictions against existing experimental data and newly acquired results.

Conclusions:

• TemStaPro provides an efficient and accurate method for predicting protein thermostability.
• The tool aids in understanding protein behavior across different temperature ranges.
• The software and data are publicly available for research use.