Modeling chimeric antigen receptor response at the single-cell level with conditional optimal transport.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Chimeric antigen receptor (CAR) T cell therapy shows promise for cancer treatment but faces challenges in clinical efficacy.
• Current computational models struggle to capture cellular heterogeneity and generalize to new CAR variants.

Purpose of the Study:

• To develop a computational framework for predicting CAR T cell therapy responses at the single-cell level.
• To enable the rational design of novel CARs by elucidating CAR design-function relationships.

Main Methods:

• An optimal transport (OT)-based framework was developed to predict CAR expression responses.
• Protein language models were used to embed CARs, extending the framework to a conditional OT-based model.
• The model was evaluated for its ability to capture gene expression changes and generalize to unseen CAR variants.

Main Results:

• The OT-based model accurately captures gene expression changes across diverse CAR variants.
• The model outperforms baseline methods for in-distribution CARs and reflects biological characteristics.
• The conditional OT-based model successfully generalizes predictions to out-of-distribution CAR designs.

Conclusions:

• Optimal transport-based modeling is valuable for understanding CAR design-function relationships.
• This framework facilitates the rational design of novel CARs with therapeutic potential.
• The developed model enhances the prediction of CAR T cell therapy responses.