InteracTor: Feature engineering and explainable AI for profiling protein structure-interaction-function...

Area of Science:

• Biochemistry and structural biology
• Computational biology and bioinformatics
• Artificial intelligence in life sciences

Background:

• Understanding protein structural and functional diversity is crucial for biology.
• Traditional methods analyzing protein structures may miss complex interactions.
• Protein interactions play a key role in biological functions.

Purpose of the Study:

• Introduce InteracTor, a novel toolkit for extracting multimodal features from protein 3D structures.
• Integrate eXplainable Artificial Intelligence (XAI) to quantify feature importance for protein classification.
• Provide mechanistic insights into protein structure, function, and dynamics.

Main Methods:

• Developed InteracTor to extract interatomic interaction features (e.g., hydrogen bonds, van der Waals forces, hydrophobic contacts) from protein 3D structures.
• Applied XAI techniques to assess the predictive power of these features in machine learning models.
• Compared the performance of interaction features against primary and secondary structure-based features.

Main Results:

• Interatomic interaction features extracted by InteracTor showed superior predictive power for protein family classification.
• The toolkit's interpretable features offer mechanistic insights into protein determinants.
• XAI integration provides transparency in assessing the predictive power of features.

Conclusions:

• Considering specific tertiary contacts is vital for accurate computational protein analysis.
• InteracTor offers a robust framework for enhancing protein function prediction models.
• This approach has significant implications for future drug discovery efforts.