Iterative Machine Learning for Classification and Discovery of Single-Molecule Unfolding Trajectories from Force...

Area of Science:

• Biophysics
• Computational Biology
• Machine Learning

Background:

• Protein unfolding analysis is crucial for understanding protein dynamics and function.
• Traditional methods for analyzing force spectroscopy data can be time-consuming and subjective.
• Classifying complex unfolding trajectories requires robust and efficient computational approaches.

Purpose of the Study:

• To develop and validate a machine learning workflow, FUSION, for rapid classification and analysis of protein unfolding trajectories.
• To expedite the interpretation of force spectroscopy data in protein biophysics.
• To identify novel unfolding pathways through automated data analysis.

Main Methods:

• Application of kernel methods, logistic regression, and triplet loss for supervised classification.
• Development of the Forced Unfolding and Supervised Iterative Online (FUSION) learning workflow.
• Validation using synthetic data generated via Monte Carlo simulations and experimental atomic force spectroscopy data.

Main Results:

• FUSION efficiently distinguished usable from unusable force spectroscopy traces.
• The workflow achieved high accuracy in classifying protein unfolding curves.
• FUSION identified previously undetected unfolding pathways in a multidomain XMod-Dockerin/Cohesin complex.

Conclusions:

• Machine learning, specifically the FUSION workflow, significantly accelerates the analysis of protein unfolding data.
• This approach enhances the ability to extract meaningful biophysical insights from complex datasets.
• FUSION demonstrates the potential of AI to advance protein biophysics research.