This was a multi-institutional collaboration to identify susceptibility mechanisms for ovarian cancer risk using gene expression and splicing data from >2,000 samples. In addition to identifying many potential risk genes, we found several loci where the disease association was correlated with splicing but not total expression, and would have been missed in a traditional TWAS. For one gene – CHMP4C – which had previously been hypothesized to be driven by a non-synonymous coding change or a nearby non-coding variant, our model implicated a splice junction variant and we showed experimentally that the risk variant was associated with differential exon usage. Overall, genes that were implicated through splicing appeared to be more essential in functional screens than those implicated through overall expression activity. As in other recent studies (see this post) many of the risk associations were observed using gene expression from tumors. The extent to which tumors can be the “right” tissue for studying risk remains a question of great interest.