Meta-analysis of massive parallel reporter assay enables functional regulatory elements prediction

Deciphering the potential of non-coding loci to influence the regulation of nearby genes has been the subject of intense research, with important implications in understanding the genetic underpinnings of human diseases. Massively parallel reporter assays (MPRAs) can measure the activity of thousands of regulatory DNA sequences and their variants in a single experiment. With the increase in the number of MPRA datasets that are publically available, one can now develop functional-based models which, given a DNA sequence, predict its regulatory activity. Here we performed a comprehensive meta-analysis of several MPRA datasets in a variety of cellular contexts. We first applied an ensemble of methods to accurately predict the MPRA output in each context, and observed that the most predictive features are consistent across data sets. We then demonstrate that predictive models trained in one cellular context can be used to accurately predict MPRA output in another, with mild loss of accuracy attributed to features that depend on the cell type. Finally, we study the extent to which MPRA can assist with the identification of genetic modifications that are associated with transcriptional changes to nearby genes. Our analysis provides insight into how MPRA data can be leveraged to highlight functional regulatory regions throughout the genome and can guide efficient design of future MPRA experiments.