A robust and interpretable end-to-end deep learning model for cytometry data.

Area of Science:

• Immunology
• Computational Biology
• Machine Learning

Background:

• Cytometry technologies are crucial for high-throughput single-cell immune analysis in immunology.
• Traditional methods like manual or automated gating can lead to significant information loss from cytometry data.
• There is a need for advanced analytical approaches to fully leverage complex cytometry datasets.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) for end-to-end analysis of cytometry data.
• To directly associate raw cytometry measurements with clinical outcomes, such as latent cytomegalovirus (CMV) infection.
• To provide an interpretable deep learning framework for cytometry data analysis.

Main Methods:

• A deep convolutional neural network (CNN) model was designed for direct analysis of raw cytometry data.
• The model was trained and tested using nine large cytometry by time-of-flight mass spectrometry (CyTOF) datasets from the ImmPort database.
• A permutation-based method was developed for interpreting the CNN model's predictions.

Main Results:

• The deep learning model accurately diagnosed latent cytomegalovirus (CMV) infection in healthy individuals using heterogeneous CyTOF data.
• The model demonstrated robustness across different studies, indicating its generalizability.
• Interpretation of the model identified a specific CD27- CD94+ CD8+ T cell population associated with latent CMV infection.

Conclusions:

• Deep learning offers a powerful, end-to-end approach for analyzing complex cytometry data, minimizing information loss.
• The developed CNN model effectively detects latent CMV infection and provides interpretable insights into immune cell populations.
• A publicly available tutorial facilitates the application of this deep learning framework for cytometry data analysis.