Using ensembles of networks to robustly identify sex-specific interactions and their associated genes. (A) A cartoon summary of how we use PANDA to build ensembles of networks using a jack-knifing approach to resample the original expression data multiple types. (B-D) Volcano plots of the difference in mean edge weight across two ensembles of networks compared to the p-value of the difference in the edge weight distributions in the two ensembles. Comparisons include (B) female versus male sputum networks, (C) female versus male blood networks, (D) female versus male “random” networks. Edges identified as “different” in each comparison are shown as either pink (female-specific) or blue (male-specific). (E-G) Venn diagrams showing the overlap in genes targeted by the female-specific (pink) or male-specific (blue) edges. Note that a gene can be targeted by both a male-specific and a female-specific edge, but by different upstream transcription factors. There is a high level of overlap in the genes targeted by the identified sex-specific edges in both the sputum and blood networks. (H-J) A hypergeometric probability was used to determine the significance of overlap in male-specific genes with genes annotated to GO categories, and female-specific genes with genes annotated to GO categories. The top five categories enriched in the males and females for each comparison are shown.