Understanding TCR T cell knockout behavior using interpretable machine learning.

Area of Science:

• Immunology
• Bioengineering
• Artificial Intelligence

Background:

• Genetic perturbation of T cell receptor (TCR) T cells is crucial for advancing cancer immunotherapies.
• Existing methods for assessing T cell responses, like cytokine assays, have limitations.
• Live-cell imaging offers a cost-effective way to observe T cell-cancer interactions, but current analysis methods are basic.

Purpose of the Study:

• To characterize T cell behavior changes induced by genetic perturbations using live-cell imaging.
• To apply explainable artificial intelligence (AI) to identify specific interaction patterns in T cell responses.
• To differentiate the effects of CRISPR knockouts (CUL5, RASA2) on T cell-cancer cell interactions.

Main Methods:

• Utilized live-cell imaging to capture T cell-cancer cell interactions over time.
• Trained convolutional neural networks (CNNs) to analyze imaging data from T cells with specific genetic modifications (CUL5, RASA2 knockouts).
• Employed explainable AI techniques to interpret CNN findings and identify behavioral differences.

Main Results:

• T cell and cancer cell coverage over time effectively marked general T cell modifications.
• Distinct cell aggregation patterns were identified as key differentiators for CUL5 knockout and RASA2 knockout T cells.
• Explainable AI successfully linked specific interaction types to different genetic perturbation conditions.

Conclusions:

• Explainable AI combined with live-cell imaging provides a powerful pipeline for analyzing complex T cell behaviors.
• Cellular aggregation is a significant behavioral marker for specific genetic perturbations in T cells.
• This methodology can be broadly applied to characterize diverse live-cell imaging datasets for improved understanding of cellular dynamics.