Abstract 1624: Antigen-independent de novo prediction of cancer-associated TCR repertoire

Abstract Cancer-associated T cells play a critical role in mediating immune responses in the anti-tumor immunity. However, due to the complex nature of cancer antigens, and the limited experimental approaches for collecting antigen-specific T cells, it remains a difficult task in cancer immunology to detect cancer-associated T cells. In this project, we have developed an antigen-independent machine learning approach to perform de novo prediction of cancer-associated T cells. Specifically, we have discovered a signature in the CDR3 sequence that can distinguish cancer-associated T cells from others. We applied this signature to immune repertoire data to define a “cancer score”, which is related to the probability of a repertoire being cancer-associated. We applied this approach to investigate 15 sample cohorts from public domain. These cohorts cover healthy donors, viral infected individuals and cancer patients from both early and late stages. Surprisingly, we observed consistently and significantly higher cancer scores using the cancer patients’ immune repertoire data, while none of the non-cancer repertoire was significant compared to healthy donors. We therefore used repertoire cancer score as a single predictor for cancer status to distinguish cancer patients from healthy donors, and observed high prediction power measured by area under the ROC (AUROC) curves. The AUROC reached 0.90 for early breast cancer patients, which is better than a number of current early prediction methods based on cancer biomarkers, such as PSA, CA-125, CEA, etc. In addition, cancer scores derived from certain late-stage cancers are associated with patient response to checkpoint blockade therapies, suggesting that it may also be used in combination with existing biomarkers, such as PD-L1 expression or cancer neoantigen load, to improve the prediction of clinical outcome of these cancer types. We anticipate broad utilities of cancer scores in cancer diagnosis and immunotherapy prognosis with the rapidly accumulating TCR repertoire sequencing data in clinical studies. Note: This abstract was not presented at the meeting. Citation Format: Bo Li. Antigen-independent de novo prediction of cancer-associated TCR repertoire [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1624.