Deep learning reveals predictive sequence concepts within immune repertoires to immunotherapy.

Area of Science:

• Immunology
• Computational Biology
• Oncology

Background:

• T cell receptor (TCR) sequencing is crucial for understanding cancer immunity.
• Current TCR analyses often overlook complementarity-determining region 3 (CDR3) sequence details, focusing on quantitative measures like clonality.

Purpose of the Study:

• To develop and apply a deep learning framework, DeepTCR, for predicting immunotherapy response.
• To leverage TCR CDR3 sequences to identify predictive signatures of treatment outcomes.
• To infer antigenic specificities and dynamic changes of predictive TCR sequences during therapy.

Main Methods:

• Utilized DeepTCR, a deep learning framework, to analyze TCR sequences.
• Identified sequence concepts predictive of immunotherapy response.
• Inferred antigenic specificities and dynamic TCR sequence changes in responders versus nonresponders.

Main Results:

• DeepTCR successfully predicted immunotherapy response.
• A predictive signature of nonresponse was associated with high frequencies of TCRs recognizing tumor-specific antigens.
• Tumor-specific TCRs exhibited greater dynamic changes during therapy in nonresponders compared to responders.

Conclusions:

• DeepTCR offers a novel approach to predict cancer immunotherapy response using TCR sequence data.
• Nonresponse is linked to an accumulation of dynamic, potentially dysfunctional, tumor-specific T cells.
• Understanding TCR sequence dynamics provides insights into treatment failure mechanisms.