Genetic Ancestry Inference From Tumor Profiling Data of 100,000 Cancer Patients

Background/Objectives: The incidence and mortality of cancer vary widely across race and ethnicity due to a variety of factors. Cancer research has underrepresented non-White patients, thereby limiting our understanding of cancer biology in these populations. Given the missingness of race/ethnicity annotations in real-word (RW) data, inferring genetic ancestry from tumor sequencing can provide a more accurate substrate to investigate such disparities in this setting.

Methods: We inferred genetic ancestry from 100,000 de-identified records from cancer patients who underwent tumor genomic profiling with the Tempus xT next-generation sequencing assay (targeting 648 genes). We used ancestry informative markers overlapping assay capture regions to infer continental ancestry proportions: Africa, Amerindian, Europe, East Asia, and South Asia. Recognizing the complexity of ancestry and race relationships, we also imputed race/ethnicity categories using admixture thresholds based on literature.

Results: While most patients in our dataset are of European descent (72%), our RW cohort includes proportionally 4.7 and 3.8-fold more patients with substantial (>50%) African and Amerindian ancestry, correspondingly, compared with TCGA. We observed higher percentages of African ancestry patients with prostate, breast, and colorectal cancer (1.8-3.1%) and Amerindian ancestry patients with colorectal cancer (2.4%) compared to the overall cohort-level distributions (p < 0.05). Using imputation on subjects lacking race/ethnicity labels, we identified 60% and 121% more patients as likely Black and Hispanic/Latino, respectively.

Conclusion: Our results show that genetic ancestry inference from tumor profiling data can partially compensate for the missingness of race/ethnicity in RW data and allow research on biological race differences in cancer etiology and outcomes.