Predicting epistasis across proteins by structural logic

Area of Science:

• Genetics and Molecular Biology
• Computational Biology
• Biochemistry

Background:

• Predicting phenotypic outcomes of genetic variations is crucial for precision medicine.
• Epistatic interactions, particularly positive epistasis like intragenic complementation, complicate these predictions.
• Intragenic complementation involves pairs of loss-of-function variants restoring protein function.

Purpose of the Study:

• To investigate intragenic complementation in the human argininosuccinate lyase (ASL) enzyme.
• To uncover the structural basis of intragenic complementation.
• To develop a predictive model for intragenic complementation using machine learning.

Main Methods:

• Utilized mutational scanning in yeast to identify intragenic complementation interactions in ASL.
• Employed machine learning algorithms leveraging protein language model embeddings.
• Validated the model's accuracy and generalizability to related enzymes like fumarase.

Main Results:

• Identified thousands of intragenic complementation interactions in ASL.
• Determined that active site assembly, not amino acid properties, drives functional restoration.
• Achieved 99.6% prediction accuracy for intragenic complementation in ASL.
• Demonstrated over 90% accuracy when generalizing the model to fumarase.

Conclusions:

• Intragenic complementation has a structural basis related to active site assembly.
• A machine learning framework can accurately predict intragenic complementation.
• This predictive framework has potential applications for at least 4% of human proteins, advancing precision medicine.