Comparing self-reported race and genetic ancestry for identifying potential differentially methylated sites in endometrial cancer: insights from African ancestry proportions using machine learning models.

Abstract

While the incidence of endometrial cancer is increasing among all US women, Black women face higher mortality rates. The reasons for this remain unclear. In this study, whole genome differential methylation analysis, along with state-of-the-art computational methods such as the recursive feature elimination technique and supervised/unsupervised machine learning models, was used to identify 38 epigenetic signature genes (ESGs) and four core-ESGs (cg19933311: TRPC5; cg09651654: APOBEC1; cg27299712: PLEKHG5; cg03150409: WHSC1) in endometrial tumors from Black and White women, incorporating genetic ancestry estimation. Methylation at two Core-ESGs, namely APOBEC1 and PLEKHG5, showed statistically significant overall survival differences between the two ancestral groups (Likelihood ratio test; P value = 0.006). Moreover, our comprehensive ancestry-based analysis revealed that tumors from women with high African ancestry exhibited increased hypomethylation compared to those with low African ancestry. These hypomethylated genes were enriched in drug metabolism pathways, indicating a potential link between genetic ancestry, epigenetic modifications, and pharmacogenomic responses. Combining ancestry, race, and disease type may help identify which patient groups will benefit most from these biomarkers for targeted treatments.