Dept. of Medical Informatics and Clinical Epidemiology.
Oregon Health & Science University
In recent years, there has been a significant academic drive to determine the cause of the missing heritability problem which is commonly associated with the genetics of complex traits such as complex diseases. In this paper, I present an in-depth literature review by introducing the nature of this problem and discussing the current state of genetic association studies which commonly involves multi-locus models as predictors (as opposed to the conventional single-locus association studies) that became prominent with the advent of genome-wide sequencing technology. Such genotype-phenotype models that involve sets of genes as predictors is wrought with analytical challenges, such as the curse of dimensionality, heterogeneity, and small main effect sizes. Machine learning as a general method to aid in the search for the missing heritability shows promise as exemplified by numerous studies by other authors, presented in this paper, demonstrating the utility of machine learning methods for addressing the specific challenges that exist in modern genetic association studies. Within the context of genetic association studies, I introduce machine learning as a general concept, present a basic overview for several different machine learning methods, and go on to present an expansive discussion on neural networks and multifactor dimensionality reduction.
School of Medicine
Jenson, Peter, "Epistasis, polygenic effects, and the missing heritability problem : a review of machine learning as applied to genetic association studies" (2015). Scholar Archive. 3626.