CytoSet

CytoSet predicts clinical outcomes from single-cell flow and mass cytometry measurements by applying permutation-invariant deep learning to patient-level cellular profiles to capture cellular heterogeneity and protein expression patterns.

Key Features:

• Deep Learning Model: Employs a deep learning framework that directly predicts patient outcomes from cellular measurements collected by flow and mass cytometry from blood or tissue samples, including datasets measuring multiple proteins across millions of cells in multi-patient cohorts.
• Permutation Invariant Architecture: Treats the profiled cells in each patient sample as an unordered set using permutation-invariant architectures to model set-structured single-cell data.
• State-of-the-Art Performance: Demonstrates superior classification performance across flow and mass cytometry benchmark datasets.
• Robustness and Scalability: Exhibits robustness to variations in the number of sub-sampled cells per patient and to model depth, enabling scalable analysis of cohorts with hundreds of patient samples.

Scientific Applications:

• Clinical prognosis and treatment stratification: Predicts patient outcomes from cytometry-derived protein expression patterns to inform prognosis and potential treatment strategies.
• Cellular heterogeneity analysis in multi-patient cohorts: Enables characterization of cellular heterogeneity across millions of single-cell profiles from flow and mass cytometry datasets.

Methodology:

Training deep learning models with permutation-invariant architectures that treat each patient's cellular profile as an unordered set to accommodate variability in cell counts and composition across samples.