Abstract Introduction: NGS-based tests that include biomarkers such as microsatellite instability (MSI) and tumor mutational burden (TMB) increase patient access to immune checkpoint inhibitors and improve the personalization of cancer care. Tumor-only NGS tests of biomarkers typically rely on germline reference sets in order to subtract germline variants from biomarker scores. Many germline reference sets lack sufficient population diversity to ensure consistent test quality for all patients. Informatic strategies such as somatic/germline zygosity prediction are often necessary to reduce bias introduced from germline variation and population-specific artefacts and mitigate the risk of inflating TMB and MSI scores. Methods: We sequenced germline samples from a curated set of research-consented patients representing diverse ancestries on a hybrid capture panel (N=1047 patients, from 10 populations). The tested groups included: African-American=416, Hispanic=162, Ashkenazi=130, Asian=127, Caucasian=85, Middle Eastern=44, Native American=27, Pacific Islander=25, Indian=25, and African=6. We confirmed that both rare population-specific alleles and population-specific technical artefacts contributed to spurious results from biomarker score inflation in underrepresented populations. To mitigate bias and universally improve scores in both MSI and TMB, we combined 1) an ancestry-diverse panel of germline samples assembled from the targeted panel, 2) population-specific public database filtering, 3) informatic germline variant subtraction, and 4) a heuristic for population-neutral target space selection. Results: Inter-population comparisons of MSI scores (Kruskal-Wallis p of 4.6*10^-69) and TMB scores (Kruskal-Wallis p of 3.7*10^-6) differed significantly in mutation signature errors prior to the mitigation work described in the methods. Application of the germline knowledge base filters and population-neutral target selection reduced the overall error rate (51% for TMB and 16% for MSI), and reduced inter-population score differences. TMB error rates reduced most for African, Pacific Islander, and African-American patients (100%, 71%, and 58%, respectively). Conversely, MSI error rates reduced most for Caucasian, Hispanic, and Asian patients (24%, 21%, and 20%, respectively). Conclusion: The incorporation of a diverse population set in this study empowered development of a more equitable bioinformatic approach, and significantly improved TMB and MSI determination across all populations. Qualitative differences in the sources of error per population and per biomarker type were associated with differing responses to each mitigation strategy. These findings underscore the technical and clinical value of incorporating diverse ancestry reference sets and identifying non-biased assay targets to optimize personalized therapy selection. Citation Format: Elizabeth R. Starks, Victoria Carlton, Veena Rajaraman, Katya Kosheleva, Tamsen Dunn, Nick Kamps-Hughes, Laurie Gay, Meaghan Russell, Sarah Albritton, John J. Vincent, Nhu Ngo. Increasing equity in MSI and TMB biomarker testing with diverse datasets and personalized informatics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1230.