Deconvolving sequence features that discriminate between overlapping regulatory annotations

Abstract

Genomic loci with regulatory potential can be annotated with various properties. For example, genomic sites bound by a given transcription factor (TF) can be divided according to whether they are proximal or distal to known promoters. Sites can be further labeled according to the cell types and conditions in which they are active. Given such a collection of labeled sites, it is natural to ask what sequence features are associated with each annotation label. However, discovering such label-specific sequence features is often confounded by overlaps between the labels; e.g. if regulatory sites specific to a given cell type are also more likely to be promoter-proximal, it is difficult to assess whether motifs identified in that set of sites are associated with the cell type or associated with promoters. In order to meet this challenge, we developed SeqUnwinder, a principled approach to deconvolving interpretable discriminative sequence features associated with overlapping annotation labels. We demonstrate the novel analysis abilities of SeqUnwinder using three examples. Firstly, SeqUnwinder is able to unravel sequence features associated with the dynamic binding behavior of TFs during motor neuron programming from features associated with chromatin state in the initial embryonic stem cells. Secondly, we characterize distinct sequence properties of multi-condition and cell-specific TF binding sites after controlling for uneven associations with promoter proximity. Finally, we demonstrate the scalability of SeqUnwinder to discover cell-specific sequence features from over one hundred thousand genomic loci that display DNase I hypersensitivity in one or more ENCODE cell lines.

An integrative view of the regulatory and transcriptional landscapes in mouse hematopoiesis

“…This systematic analysis of transitions in epigenetic states across cell types helps uncover the differentiation history of cCREs and provides mechanistic insights into regulation. For example, by using SeqUnwinder (Kakumanu et al 2017) to discover discriminative motifs, we found that the CMP cCREs that transition from poised to active enhancer in the erythroid lineage were enriched for the GATA transcription factor binding site motif, whereas those that transition to a polycomb state were enriched in motifs for binding ETS transcription factors such as SPI1 (also known as PU.1) (Supplemental Fig. S13).…”

An integrative view of the regulatory and transcriptional landscapes in mouse hematopoiesis

“…This systematic analysis of transitions in epigenetic states across cell types helps uncover the differentiation history of cCREs and provides mechanistic insights into regulation. For example, using SeqUnwinder (Kakumanu et al 2017) to discover discriminative motifs, we found that the CMP cCREs that transition from poised to active enhancer in the erythroid lineage were enriched for the GATA transcription factor binding site motif, whereas those that transition to a polycomb state were enriched in motifs for binding ETS transcription factors such as PU.1 (Supplemental Fig. S14).…”

An interpretable bimodal neural network characterizes the sequence and preexisting chromatin predictors of induced TF binding

“…To overcome the local behavior of such attribution methods, we extracted 20 bp sequences surrounding the local IG peaks; i.e., we extracted regions or "hills" that drive the neural network output from each sequence bound by Ascl1. We clustered the IG-derived hills based on their underlying sequence features using a K-means based clustering procedure described previously in SeqUnwinder 69 . Briefly, only k-mers present in at least 5% of IG-derived hills are used for clustering, and K-means clustering is performed using Euclidean distance as a metric 69 .…”

“…We clustered the IG-derived hills based on their underlying sequence features using a K-means based clustering procedure described previously in SeqUnwinder 69 . Briefly, only k-mers present in at least 5% of IG-derived hills are used for clustering, and K-means clustering is performed using Euclidean distance as a metric 69 . We then perform motif discovery using MEME 70 to identify the enriched motifs within each K-means defined cluster 69 .…”

“…Briefly, only k-mers present in at least 5% of IG-derived hills are used for clustering, and K-means clustering is performed using Euclidean distance as a metric 69 . We then perform motif discovery using MEME 70 to identify the enriched motifs within each K-means defined cluster 69 .…”