Written by: Stephen Hsu
Primary Source: Information Processing
This is an ASHG poster (click for larger version) describing work on predictive modeling of genetic disease using Compressed Sensing. Our previous work dealt with continuous traits (quantitative phenotypes). In the case of disease, one sometimes only has binary data to work with: individuals in the sample are either cases (have the condition) or controls (do not have the condition). Their underlying genetic susceptibility to the condition is not directly measurable. However, sophisticated techniques can use even this type of data to deduce the underlying genetic architecture. As in our earlier work, we demonstrate a “phase transition” in the performance of our algorithms as the amount of data available increases.
See related posts on quantitative traits: linear models 2, linear models, nonlinear method, and this talk: Genetic architecture and predictive modeling of quantitative traits.