CyTOF

CyTOF applies an end-to-end deep convolutional neural network to analyze mass cytometry (CyTOF) single-cell measurements and associate raw cytometry data with clinical outcomes such as latent cytomegalovirus (CMV) infection.

Key Features:

• Deep Convolutional Neural Network: Processes raw CyTOF data directly in an end-to-end architecture for single-cell analysis.
• Clinical Outcome Association: Links raw cytometry measurements with clinical outcomes, exemplified by association with latent CMV infection.
• Data Heterogeneity Handling: Demonstrated on nine large CyTOF studies from the ImmPort database, enabling diagnosis of latent CMV across heterogeneous study cohorts.
• Permutation-Based Interpretability: Employs a permutation-based method to interpret model outputs and identify associations between features and outcomes.
• Identification of Cell Subsets: Identifies significant cell populations such as CD27- CD94+ CD8+ T cells associated with latent CMV infection.

Scientific Applications:

• Immunology: Analysis of high-throughput single-cell immune measurements to study immune responses and latent infections such as CMV.
• Biomarker Discovery: Identification of immune cell populations and signatures associated with clinical outcomes for downstream investigation.

Methodology:

Training a deep convolutional neural network using Keras and TensorFlow on large-scale CyTOF datasets and interpreting results with a permutation-based method.