Leveraging the Microbiome, Local Admixture, and Machine Learning to Optimize Anticoagulant Pharmacogenomics in Medically Underserved Patients

点击翻译按钮获取中文摘要

ABSTRACT Warfarin remains one of the most commonly prescribed drugs and a leading cause of emergency hospitalizations. Warfarin use is especially common in medically underserved patients such as African Americans (AAs) and Latinos, which is particularly concerning since AAs and Latinos suffer worse outcomes due to suboptimal warfarin therapy. Thus AAs and Latinos can derive a distinct benefit from warfarin pharmacogenomic (PGx) algorithms, which maximize safety and efficacy by predicting individualized warfarin dose. However, currently available PGx algorithms have critical limitations, including a lack of generalizability to non-white populations and a failure to account for 50 percent of variability in warfarin dose. Under-representation in clinical studies, the propensity to cause adverse events, and a lack of consideration of admixed populations in clinical PGx guidelines are all factors that contribute to limited utility of warfarin PGx algorithms in diverse populations. Many potential sources of warfarin stable dose variability remain critically unexplored, including the role of vitamin K biosynthesizing bacterial species, the influence of local ancestry at warfarin pharmacogenes, and the potential for machine learning techniques to enable accurate warfarin dosing algorithms in diverse populations. This proposal addresses the overarching hypothesis that warfarin stable dose prediction can be improved by incorporation of gut microbiome data, measures of local ancestry, and machine learning in diverse populations. We will pursue three Specific Aims (SAs) to test this hypothesis: (SA1) Determine the impact of abundance of vitamin K biosynthesizing bacteria from the gut microbiome on warfarin stable dose and; (SA2) Determine the influence of local admixture on warfarin stable dose in admixed populations; (SA3) Optimize warfarin PGx algorithms for diverse populations using machine learning. In SA#1, we will conduct a clinical study with fecal sample collection at anticoagulation clinic visits and perform whole genome bacterial sequencing to identify the effect of vitamin K biosynthesizing bacterial species on warfarin stable dose. In SA#2, we will estimate African, European, and Native American local ancestry in warfarin pharmacogenes in a large, admixed population (n=1194) and determine its effects on warfarin stable dose. In SA#3, a large, diverse population of warfarin treated patients (n=7366) will be used to develop machine learning models and test improved prediction of warfarin stable dose over existing linear regression models. Our studies overcome major limitations of previous warfarin PGx studies by leveraging gut microbiome data, local ancestry, machine learning, and diverse, admixed populations. The outcomes of this work will provide a framework for local ancestry investigation with other PGx drug-gene pairs, enabling use of clinical PGx guidelines in admixed populations. This research has the potential to identify new sources of variability in warfarin dose, improve the safety and efficacy of warfarin treatment, and reduce disparities in PGx research for medically underserved patients.