Predicting Functional Surface Topographies Combining Topological Data Analysis and Deep Learning Across the Human...

Area of Science:

• Structural biology
• Bioinformatics
• Computational biology

Background:

• Protein structure characterization is crucial for understanding biological function.
• AI tools like AlphaFold2 (AF2) have significantly advanced protein structure prediction.
• A gap exists in functional annotations and surface topographical information for predicted protein structures.

Purpose of the Study:

• To develop a method for predicting functional pockets in AI-predicted human protein structures.
• To associate predicted functional pockets with Gene Ontology (GO) terms and Enzyme Commission (EC) numbers.
• To provide a comprehensive resource of functional pocket predictions for human proteins.

Main Methods:

• Combined topological data analysis with the deep learning method DeepFRI.
• Applied the method to a dataset of 65,013 AlphaFold2-predicted human representative protein structures.
• Mapped predictions to 186,095 non-fragment AlphaFold2-predicted human protein structures.

Main Results:

• Successfully predicted functional pockets for a large set of human protein structures.
• Associated predicted pockets with relevant GO terms and EC numbers.
• Made all predicted functional pocket data publicly accessible online.

Conclusions:

• The developed method effectively predicts functional pockets and their annotations in AF2-predicted human proteins.
• This resource enhances the functional understanding of predicted protein structures.
• The findings facilitate further research in protein function and drug discovery.