Discovery of Novel Gain-of-Function Mutations Guided by Structure-Based Deep Learning.

Area of Science:

• Biochemistry
• Computational Biology
• Protein Engineering

Background:

• Deep learning shows promise for accelerating protein engineering, but practical applications and examples of improved proteins remain limited.
• Existing methods often rely on energetics-based approaches, which may not capture all factors contributing to protein function.
• Understanding the chemical microenvironments of amino acids is crucial for predicting functional mutations.

Purpose of the Study:

• To develop and validate a deep learning model for identifying novel gain-of-function mutations in proteins.
• To demonstrate that the model can identify mutations not predicted by traditional energetics-based methods.
• To explore the chemical space within protein microenvironments to understand mutation-induced functional changes.

Main Methods:

• A 3D convolutional neural network was trained to correlate amino acids with their neighboring chemical microenvironments.
• The model was used to guide the identification of novel gain-of-function mutations.
• The identified mutations were combined and tested for their effect on protein function *in vivo* across three different proteins.

Main Results:

• The deep learning model successfully identified novel gain-of-function mutations missed by energetics-based approaches.
• Combining these mutations resulted in a significant improvement of protein function *in vivo*, with at least a 5-fold increase across three diverse proteins.
• The model enabled the interrogation of chemical space, revealing specific interactions responsible for gain-of-function phenotypes.

Conclusions:

• Deep learning, specifically 3D convolutional neural networks, can effectively guide the discovery of beneficial protein mutations.
• This approach offers a powerful alternative to traditional methods, expanding the possibilities in protein engineering.
• The model provides insights into the chemical basis of protein function and mutation effects.