Cell morphology-based machine learning models for human cell state classification.

Area of Science:

• Computational Biology
• Cell Biology
• Biotechnology

Background:

• Flow cytometry is crucial for cell analysis.
• Distinguishing apoptotic cells traditionally requires fluorescent stains.
• Cell size and granularity are key flow cytometry parameters.

Purpose of the Study:

• To develop machine learning models for classifying healthy versus apoptotic cells.
• To utilize only forward (FSC) and side (SSC) scatter data, excluding fluorescent markers.
• To create a stain-free, automated cell classification method.

Main Methods:

• Trained machine learning classifiers (logistic regression, random forest, k-NN, MLP, SVM) using FSC/SSC derived features (area, height, width).
• Used Annexin V/propidium iodide staining to label apoptotic cells for training data generation.
• Evaluated 1046 candidate models on their ability to predict cell states.

Main Results:

• A multilayer perceptron model achieved high performance metrics (0.91 precision, 0.93 recall, 0.92 F1-score, 0.97 AUC).
• The model successfully classified cells using only size and granularity information.
• Performance was compared to traditional scatter gating techniques.

Conclusions:

• Machine learning models can reliably classify apoptotic cells using only FSC/SSC data.
• This stain-free method provides an automated and efficient alternative to traditional staining assays.
• The developed module is applicable to diverse flow cytometry analyses.