A variational deep-learning approach to modeling memory T cell dynamics.

Area of Science:

• Immunology
• Computational Biology
• Systems Biology

Background:

• Mechanistic models traditionally simplify immune cell populations using limited markers.
• High-dimensional single-cell data presents challenges for existing modeling approaches.
• Understanding tissue-resident memory T cell (T_RM) dynamics is crucial for effective immunity.

Purpose of the Study:

• To develop a novel computational approach for analyzing high-dimensional single-cell data in immune responses.
• To investigate the dynamics and heterogeneity of lung CD4 and CD8 T_RM cells during influenza infection resolution.
• To provide a framework for interpreting complex, time-series, high-dimensional biological data.

Main Methods:

• Developed a deep learning framework integrating stochastic variational inference.
• Trained the model directly on single-cell flow cytometry data, inferring population structure and model parameters simultaneously.
• Applied the method to study lung T_RM development and persistence in a mouse influenza infection model.

Main Results:

• Identified significant phenotypic diversity within lung CD4 and CD8 T_RM populations during immune response resolution.
• Demonstrated distinct, time-dependent dynamics among T_RM subsets.
• Revealed that long-term T_RM heterogeneity is sustained by differentiation from persistent Bcl-2^hi subsets.

Conclusions:

• The study provides novel insights into the complex dynamics of tissue-localized immune memory.
• The developed computational approach offers a powerful new basis for interpreting high-dimensional, time-series biological data.
• This methodology is broadly applicable to various biological systems requiring analysis of complex cellular dynamics.