Hypothesis
Published computational methods for linking enhancers to their target genes would clearly beat a simple genomic-distance baseline when tested on a curated benchmark.
What failed
On the new BENGI benchmark, which combines the cCRE registry with experimentally derived interactions, the best method (TargetFinder) was only modestly better than the distance baseline within a cell type and often no better than it across cell types.
Broken assumption
Supervised enhancer-gene predictors learn regulatory signal that generalises beyond simple proximity.
Why
The abstract gives no mechanism; it concludes current methods need improvement and offers BENGI as a framework for developing them.
From the published paper: Moore et al. Genome biology, 2020. ↗CC BY