Jackknife Model Averaging Prediction Methods for Complex Phenotypes with Gene Expression Levels by Integrating External Pathway Information

Figure 1

Comparison of predictive performance of four models with JMAP with PVE = 0.3. Performance is measured by R difference with respect to JMAP; therefore, a negative value (i.e., values below the horizontal line) indicates worse performance than JMAP. In each setting, five groups with nonzero effect sizes were selected; I represents the settings where all the genes in the five groups had nonzero effect sizes; II represents the settings where only the genes in the first two groups had nonzero effect sizes, and half of the genes in the last three groups had nonzero effect sizes; III represents the settings where the effect sizes of the first two groups were nonzero, and the proportion of nonzero effect sizes in the last three groups was 80%, 50%, or 20%; IV represents the settings where the proportion of nonzero effect sizes in the five groups was 90%, 70%, 50%, 30%, or 10%. The predictive performance was assessed across 100 replicates in each scenario.