In the second example (Fig. 4, middle column), most effect estimates in non-brain tissues are positive (30 out of 34), but modest in size, and only one is nominally significant (p < 0.05). However, by combining information among tissues, mash estimated that all effects in non-brain tissues are positive, and mostly “significant” (lfsr < 0.05). By contrast, the estimated effects in brain tissues are inconsistent (both positive and negative) and so mash is not confident about the sign of effects in brain tissues. This example illustrates that mash can learn to treat subsets of conditions differently; mash learned that effects in brain tissues are occasionally different from effects in other tissues, and therefore did not draw strong inferences in the brain based on the other tissues.
