T-cell receptor (TCR) repertoire profiling has emerged as a powerful tool for biological discovery and biomarker development in cancer immunology and immunotherapy. A key statistic derived from repertoire profiling data is diversity, which summarizes the frequency distribution of TCRs within a mixed population. Despite the growing use of TCR diversity metrics in clinical trial correlative studies in oncology, their accuracy has not been validated using published ground-truth datasets. Here, we reported the performance characteristics of methods for TCR repertoire profiling from RNA-sequencing data, showed undersampling as a prominent source of bias in diversity estimates, and derived a model via statistical learning that attenuates bias to produce corrected diversity estimates. This modeled diversity improved discrimination in The Cancer Genome Atlas data and associated with survival and treatment response in patients with melanoma treated with anti-PD-1 therapy, where the commonly used diversity normalizations did not. These findings have the potential to increase our understanding of the tumor immune microenvironment and improve the accuracy of predictions of patient responses to immunotherapy.
©2020 American Association for Cancer Research.