Figure 4

Sensitivity increases with the proportion population initialized from domain knowledge. This figure illustrates the sensitivity of GENN to detect both functional SNPs as the proportion of the NN population initialized from domain knowledge increases from 0 to 99%. Panels going left to right show the increasing implication index of the model that includes both functional variables. Rows of panels show the population size per deme. The X-axis in each panel shows the proportion of the initial NN population which was seeded with two-SNP models from a domain knowledge source. Solid line shows when GE alone was used to train NNs (no BP). Dashed line shows sensitivity when using the hybrid BP-GENN algorithm (see methods). Faint horizontal solid and dashed lines show for reference the baseline sensitivity, for GENN and BP-GENN, when the population was initialized randomly, i.e. 0% initialized from domain knowledge. This figure indicates that sensitivity increases as the proportion of the NN population initialized from domain knowledge increases, and the increase is more notable in smaller population sizes.