Tiled Regression (Wilson et al., 2009) is an approach to determining a regression model of trait variation when the number of possible genetic predictors is very large. It focuses initially on moderate-sized segments of the genome called "tiles" and discards those showing no evidence of significant effect on the trait. Within the more promising tiles, stepwise or penalized regression is used to select a subset of predictors that independently contribute to trait variation. Predictors that are not discarded are then combined across tiles for selection within chromosome, and then across the genome. Quantitative traits are modeled with linear regression, and binary traits are modeled with logistic regression, in each case considering only additive effects until the final stage. Two-way interactions among predictors remaining after genome-wide selection may then be added for consideration in a final selection step. Family relationships are handled with generalized estimating equations (GEE). TRAP (Sorant et al., 2010) is a package of R functions implementing the Tiled Regression analysis method.
Last Modified: Tuesday, 06-Oct-2015 09:34:58 EDT