Whole Genome Prediction (WPG) methods make use of all markers available on a genome-wide array, even in settings where their number far exceeds that of individual study participants (p>>n settings). However, these WGP methodologies were originally developed within the context of genomic selection in animal studies, and have not been tailored to outcome prediction in randomized controlled trials (RCT) of lifestyle interventions. In particular, they assume genetic independence of the treatment and control groups, whereas RCTs draw upon a common study population. Dr. Papandonatos and his students have sought to incorporate WPG approaches into personalized treatment recommendations by accounting for the underlying genetic heterogeneity present in human populations, driven by a) hidden population stratification; b) population admixture; and c) cryptic relatedness through a correlation matrix that reflects the genetic relationship structure of the entire target population and cannot be altered by treatment. Phenotypes of interest include, but are not limited to, weight loss and weight loss maintenance, physical activity of both moderate and vigorous intensity, smoking reduction and abstinence.
Genetics and Genomics
Interpretable Machine Learning Methods for Genome-wide Association Mapping
The goal of this project is to build machine learning algorithms and statistical tools that aid in the understanding of how nonlinear interactions between genetic features affect the architecture of complex traits and contribute to disease etiology. A key theme of this work will be to take modern computational approaches and develop theory that enable their interpretations to be related back to classical genomic principles.
-
Lorin Crawford
Distinguished Senior Fellow in Biostatistics
Deep learning Methods for Fine Mapping and Discovery in Genomic Association Studies
During this time period, we will develop a suite of novel probabilistic machine learning and deep neural networks tools for fine mapping and discovery in genomic sequencing studies. Specifically, (i) develop an interpretable significance measure for probabilistic machine learning methods, (ii) develop a unified deep learning framework for gene-level and pathway enrichment analysis in genome-wide association studies, and (iii) create distributable software and use it to characterize nonlinear genetic effects at multiple genomic scales in real data applications.
-
Lorin Crawford
Distinguished Senior Fellow in Biostatistics