Knowledge Gap: Functional annotation, especially for non-coding regions, remains challenging. Current tools (CADD) does not use tissue-specific information provided by projects like ENCODE/RME/GTEx.

Strategy:

Here, they propose to integrate RME data for more than 100 cell-types to derive 7 tissue-specific predictive models (brain, GI, lung, heart, blood, muscle and epithelium), meaning that one locus might be functional in one tissue but not in others.

To ease the model fitting, they make another assumption (that I think is a bit strong): each annotation mark is conditionally independent given the functionality, and then derive a posterior probability for each locus.

Main findings:

they estimate that 22% of the genome is functional, for at least one tissue, and ~2% is functional across the 7 tissues.

On Psychiatric Genomics Consortium data (SCZ), they demonstrate a better prioritarization using brain model rather than heart, and opposite results were observed on 23 CARDIoGRAM loci.

Strengths:

UCSC track for whole-genome annotation

preprint of GenoSkyline-Plus that includes RNA-Seq information from RME too

Weaknesses:

strong assumption on the probabilistic model