Label-free macrophage phenotype classification using machine learning methods.

Area of Science:

• Immunology
• Cell Biology
• Computational Biology

Background:

• Macrophages are crucial innate immune cells with diverse phenotypes.
• Identifying macrophage phenotypes is vital for understanding immune responses.
• While markers are common, morphology and autofluorescence offer alternative classification clues.

Purpose of the Study:

• To explore macrophage autofluorescence as a distinct feature for phenotype classification.
• To develop and evaluate a machine learning framework for identifying macrophage phenotypes based on autofluorescence.
• To assess the accuracy of classifying six specific macrophage phenotypes (M0, M1, M2a, M2b, M2c, M2d).

Main Methods:

• A dataset of 152,438 cell events with 45 optical signals was constructed.
• Supervised machine learning methods were applied to analyze the autofluorescence signal fingerprints.
• A fully connected neural network was employed for classification.

Main Results:

• The fully connected neural network achieved 75.8% accuracy in classifying six simultaneous macrophage phenotypes.
• Classification accuracy increased significantly when reducing the number of phenotypes (e.g., 92.0% for two phenotypes).
• Intrinsic autofluorescence proved to be a valuable feature for macrophage phenotyping.

Conclusions:

• Macrophage autofluorescence is a viable intrinsic feature for phenotype classification.
• The proposed machine learning approach is quick, simple, and cost-effective.
• This method can accelerate the discovery of macrophage phenotypical diversity.