Atom-specific persistent homology and its application to protein flexibility analysis

Area of Science:

• Computational Biology
• Topological Data Analysis
• Structural Bioinformatics

Background:

• Persistent homology is successful in analyzing biomolecular data by examining atomic connectivity at various scales.
• However, its application to local atomic properties like flexibility and B-factor prediction is limited.

Purpose of the Study:

• To introduce atom-specific persistent homology for local atomic-level molecular representation.
• To apply this method for analyzing and predicting localized properties in macromolecules.

Main Methods:

• Developed atom-specific persistent homology using conjugated sets of atoms and simplicial complexes.
• Utilized Bottleneck and Wasserstein metrics to measure differences in topological invariants.
• Integrated atom-specific topological features with machine learning algorithms (gradient boosting trees, CNN).

Main Results:

• The proposed method provides localized atomic-level topological representations.
• Successfully applied to protein thermal fluctuation analysis and B-factor prediction.
• Demonstrated effectiveness in analyzing and predicting localized information in macromolecules.

Conclusions:

• Atom-specific persistent homology offers a powerful new topological tool for biomolecular analysis.
• This method enables the prediction of localized atomic properties previously difficult to assess.
• The approach integrates global topological tools with local property analysis effectively.