Detecting Epistasis in Living Systems
Epistasis is the phenomenon of pairwise, or higher order, interactions, which can be non-linear, between genetic loci. Classic approaches designed to detect phenotypic variation are limited in scope and generally only can detect additive genetic variation of each locus using univariate approaches. However, the genetics of most traits are likely complex and involve large biological pathways that are underlaid by dense networks of interacting genes and regulatory proteins. A major focus of my research is the development and implementation of algorithms and computational approaches to detect genetic interactions and describe these networks. These include extensions of linear models to detect complex interactions, epistatic network analyses, and unsupervised learning approaches focused on transparency and explainability. Furthermore, I am interested in exploring different ways to encode genetic interactions outside of a standard Cartesian (multiplicative) model as biological epistasis likely involves complex, non-linear relationships.
Development of AutoML Approaches for QTL analysis/GWAS
QTL analysis and GWAS have revolutionized the study of quantitative genetics. However, this approaches are limited in their ability to describe non-additive genetic variation. AutoML approaches, especially genetic programming (GP) algorithms, have the capability to explore large search spaces quickly and exhaustively. This capability, coupled with ML operators designed with quantitative genetics in mind, can detect non-additive sources of genetic variation. I am actively developing and exploring AutoML approaches that make the kinds of decisions a trained geneticist would make when designing experiments that attempt to associate genotypes to phenotypes while also taking into consideration epistasis, non-additive inheritance models, and expert/domain knowledge.
Check out our recent publication on our proof-of-concept, AutoQTL!
Prediction of Surgical Outcomes, Pain, and Opioid Use Disorder
A key determinant of surgical success is pain perceived by patients post-operation. Further, opioid analgesic dosage is highly associated with perceived pain levels. I currently work on a number of projects that aim to predict opioid dosage and pain as measures of surgical success. Additionally, I am interested in the clinical features that drive these predictions, which include demographics, comorbid psychiatric diagnoses (including substance use and other psychiatric disorders), pertinent aspects of the patient’s medical history, and environmental/societal factors. It is a goal of these research projects to develop a pain/opioid misuse risk score that can better inform clinicians and patients of the risks of opioid analgesic exposure and potential surgical complications.
Validating the Adaptive Decoupling Hypothesis
The adaptive decoupling hypothesis, postulated by Nancy Moran in 1994, postulates that complex life cycles (CLCs), where each phase of an organism’s ontogeny is distinct in morphology, behavior, and physiology, evolve to exploit different ecological niches. Many ecologically, medically, and agriculturally relevant animals have complex life cycles, including most insect species. I am interested in determining if distinct genetic architectures underlie adaptive decoupling for fitness-related traits. My previous research has shown a lack of statistically significant correlations and low genetic correlations for thermal tolerance phenotypes between larvae and adults in Drosophila melanogaster. Additionally, distinct transcriptomic profiles and genotype-by-environment interaction landscapes exist between these two life stages. Thus, for these traits, there appears to be little genetic constraint across ontogeny. I am planning future work that will focus on expanding my analyses to other species and phenotypes as well as exploring differences in detected epistatic events between life stages in species with CLCs.