Transfer learning to leverage larger datasets for improved prediction of protein stability changes

Area of Science:

• Biophysics
• Computational Biology
• Protein Engineering

Background:

• Amino acid mutations can decrease protein thermodynamic stability, leading to various diseases.
• Engineered proteins with improved stability are vital for biomedical research and therapeutic applications.
• Accurate computational prediction of mutation-induced stability changes is essential but challenging due to data limitations.

Approach:

• Introduced ThermoMPNN, a deep neural network for predicting protein stability changes from point mutations using structural information.
• Utilized a newly released mega-scale dataset for training a robust stability prediction model.
• Employed transfer learning with features from a protein sequence prediction network to enhance model performance.

Key Points:

• ThermoMPNN demonstrates competitive performance on benchmark datasets.
• The model employs a lightweight architecture for rapid and scalable predictions.
• Leveraged large-scale stability data and transfer learning for improved accuracy.

Conclusions:

• ThermoMPNN provides a valuable tool for protein stability prediction and protein design.
• The study highlights the utility of large datasets and transfer learning in computational protein stability analysis.
• The developed model is accessible for researchers and medical professionals.