We cannot conclusively say whether gene expression or chromatin data is preferable when both types of data are available in the same tissues and cell types (Online Methods, Figure S4, Figure S17, Table S10, Table S16). Instead, we conclude that the question of which type of data is preferable may depend on complex factors such as which chromatin marks were analyzed, the sample size with which the specifically expressed genes are called, and the overall quality of the data set. When gene expression and chromatin data are available on the same set of tissues or cell types, it may be possible to combine these types of data to improve power, for example by restricting an annotation to tissue-specific chromatin marks near specifically expressed genes, or by combining the P-values from separate analyses of the two types of data. We defer a thorough exploration of this set of possibilities to future work.
