Machine learning predictions of T cell antigen specificity from intracellular calcium dynamics

Area of Science:

• Immunology
• Bioinformatics
• Machine Learning

Background:

• Adoptive T cell therapies require T cells with specific tumor-targeting T cell receptors (TCRs).
• Identifying effective TCR sequences is a major bottleneck in TCR-engineered cell therapy production.
• Intracellular calcium fluctuations signal T cell receptor (TCR) engagement but are highly variable.

Purpose of the Study:

• To investigate the potential of machine learning algorithms for classifying T cell activation from complex calcium signaling data.
• To develop a foundation for an antigen-specific TCR sequence identification pipeline for adoptive T cell therapies.

Main Methods:

• Utilized deep learning tools to analyze intracellular calcium fluctuations as a readout of T cell receptor (TCR) signaling.
• Trained and tested a machine learning algorithm to predict T cell activation based on calcium signaling data.
• Validated the algorithm using TCR-transgenic CD8+ T cells, distinct TCRs, and polyclonal T cells.

Main Results:

• Demonstrated accurate prediction of TCR-transgenic CD8+ T cell activation using deep learning models based on calcium fluctuations.
• Successfully tested the algorithm's performance against T cells with different TCRs and in polyclonal T cell populations.
• Established the feasibility of using machine learning to interpret variable calcium signaling events for T cell classification.

Conclusions:

• Machine learning, specifically deep learning, can effectively classify T cell activation from variable calcium signaling data.
• This approach provides a promising foundation for a novel pipeline to identify antigen-specific TCR sequences for adoptive T cell therapies.
• Overcoming the bottleneck in TCR identification can accelerate the development of personalized cancer immunotherapies.