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

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Protein thermodynamic stability is crucial for biological function and implicated in diseases.
• Accurate prediction of mutation-induced stability changes is vital for protein design and medical applications.
• Existing computational methods face challenges due to limited high-quality training data.

Purpose of the Study:

• To develop a deep learning model for predicting protein point mutation stability changes.
• To demonstrate the effectiveness of large-scale stability datasets and transfer learning for model training.
• To provide a rapid and scalable tool for protein stability prediction and engineering.

Main Methods:

• Developed ThermoMPNN, a deep neural network utilizing a megascale stability dataset.
• Employed transfer learning with features from ProteinMPNN, a protein sequence prediction model.
• Validated performance on established benchmark datasets.

Main Results:

• ThermoMPNN achieves state-of-the-art performance in predicting protein stability changes.
• The model demonstrates robustness when trained on large, high-quality datasets.
• The lightweight architecture enables fast and scalable predictions.

Conclusions:

• ThermoMPNN offers a powerful and efficient tool for predicting the impact of mutations on protein stability.
• The study highlights the utility of large datasets and transfer learning in computational protein design.
• ThermoMPNN is available as an open-source tool to facilitate protein stability prediction and engineering.