Predicting phenotype from genotype
CALS Impact Statement
Overview
Other
View All
Overview
abstract
At the intersection of our work on human cardiovascular disease and our work on immune responses in Drosophila is the problem of prediction. Given genotypic attributes of flies or people, how can we best build a model that allows prediction of phenotype (disease risk, immune competence) given nothing other than the genotype data? Many individuals in the field of systems biology are trying to do this with machine-learning sorts of methods, and we wanted to instead apply knowledge of the underlying pathways to make predictions. Our work on the reverse cholesterol transport pathway has given very sobering insight into the nature of cardiovascular disease, and places limits on how well ANY model could predict this disease. In particular, about half the risk is already known to be induced by diet and exercise, and so at best genotype explains the other half. In fact, genetics explains well under half, because most of that remaining variance is explained by interactions between diet, excerise and genotype. The Drosophila data are much more amenable to this kind of modeling, and the control of the genetic variation seems to be largely why (in controlled backgrounds, single SNPs often explain 15 percent or more of the variance in immune phenotypes). In the end we hope to be able to identify which phenotypes are most easily predicted by genotypic variation, and to better understand why.
Other
submitted by
Clark, Andrew G. Jacob Gould Schurman Professor of Population Genetics
issue being addressed
At the intersection of our work on human cardiovascular disease and our work on immune responses in Drosophila is the problem of prediction. Given genotypic attributes of flies or people, how can we best build a model that allows prediction of phenotype (disease risk, immune competence) given nothing other than the genotype data? Many individuals in the field of systems biology are trying to do this with machine-learning sorts of methods, and we wanted to instead apply knowledge of the underlying pathways to make predictions. Our work on the reverse cholesterol transport pathway has given very sobering insight into the nature of cardiovascular disease, and places limits on how well ANY model could predict this disease.
response
The best use of genotypic data in medicine would be prediction. If only we knew who was particularly at risk, efforts to ameliorate that risk by behavioral, dietary, exercise or drug therapies would be truly preventitive instead of post hoc.
impact assessment
The human cardivascular efforts have resulted in many publications that mostly identify the cause of the problem in prediction. The Drosophila data are just about to turn the corner and realize the benefits of a full gene-network analysis.
Our lab has provided an environment for learning for a range from 6 to 11 postdoctoral fellows, 4-6 graduate students, 4-7 undergraduates, and two technicians.
