TDAExplore: Quantitative analysis of fluorescence microscopy images through topology-based machine learning.

Area of Science:

• Biotechnology
• Computational Biology
• Microscopy

Background:

• Machine learning advances automated information extraction from fluorescence microscopy data.
• Existing models often require extensive training datasets (hundreds to thousands of images).
• Current models lack the ability to identify image regions contributing to classification.

Purpose of the Study:

• Introduce TDAExplore, a machine learning pipeline utilizing topological data analysis (TDA).
• Enable efficient classification of cellular perturbations using limited training data.
• Provide quantitative, spatial information on image regions contributing to classification.

Main Methods:

• Developed TDAExplore, a machine learning image analysis pipeline based on topological data analysis.
• Trained models using only 20-30 high-resolution images and whole-image labels.
• Ensured robustness across multiple subjects and microscopy modes.

Main Results:

• TDAExplore successfully classifies cellular perturbations with high accuracy.
• Achieved robust performance using significantly reduced training datasets (20-30 images).
• Generated quantitative, spatial characterization of image regions critical for classification.

Conclusions:

• TDAExplore offers an accessible and powerful tool for analyzing fluorescence microscopy data.
• The pipeline requires modest computational resources, suitable for standard PCs.
• Provides valuable quantitative and spatial insights for diverse biological applications.