Apr 22, 2024, 3:00 pm4:00 pm


Event Description

Humans are related to each other in complicated ways, potentially sharing a small number of recent ancestors, a larger number of more ancient ancestors from one or various parts of the world, both, or rarely neither. What results is a high-dimensional covariance structure between the genetic variants of individuals called the population kinship matrix. While some applications fully model relatedness, many continue to assume unstructured populations, or low-dimensional approximations that suffice for multiethnic or admixed cohorts, but cannot model family and cryptic relatedness. The long term goal of the Ochoa lab is to develop models for diverse human genetic data. Our short term goals are focused on the applications of genetic association studies,  meta analysis, admixture inference, heritability estimation, polygenic risk scores, and linkage disequilibrium (LD). A key result driving much of our research is our recent development of an unbiased kinship estimator. Using theory, simulations, and real human genetic data, we have found that cryptic relatedness is a much larger confounder in human association studies than has been previously recognized. We also find that cryptic relatedness explains whether meta analysis will have inflated statistics or not. We analyzed an admixed family to illustrate how ancestry and family structure combine to result in the total population structure, and are developing a method to estimate admixture parameters from an unbiased population kinship matrix. Surprisingly, we discovered that the common kinship estimation bias is compensated for in association studies; however, this bias carries over to heritability estimation using variance components. We have also generalized the LD score regression method to work with statistics from mixed effects models and more general LD models. Lastly, we have identified a promising connection between kinship and LD that may result in improved LD estimation models. Overall, our lab is working on improving several key approaches aiming for greater accuracy and applicability for the complex human genetic datasets that are fast becoming the norm.

Event Category
QCB Seminar Series