Integrative prediction of gene expression with chromatin accessibility and conformation data

Abstract

Background: Enhancers play a fundamental role in orchestrating cell state and development. Although several methods have been developed to identify enhancers, linking them to their target genes is still an open problem. Several theories have been proposed on the functional mechanisms of enhancers, which triggered the development of various methods to infer promoter enhancer interactions (PEIs). The advancement of high-throughput techniques describing the three-dimensional organisation of the chromatin, paved the way to pinpoint long-range PEIs. Here we investigated whether including PEIs in computational models for the prediction of gene expression improves performance and interpretability. Results: We have extended our Tepic framework to include DNA contacts deduced from chromatin conformation capture experiments and compared various methods to determine PEIs using predictive modelling of gene expression from chromatin accessibility data and predicted transcription factor (TF) motif data. We found that including long-range PEIs deduced from both HiC and HiChIP data indeed improves model performance. We designed a novel machine learning approach that allows to prioritize TFs in distal loop and promoter regions with respect to their importance for gene expression regulation. Our analysis revealed a set of core TFs that are part of enhancer-promoter loops involving YY1 in different cell lines. Conclusion: We show that the integration of chromatin conformation data improves gene expression prediction, underlining the importance of enhancer looping for gene expression regulation. Our general approach can be used to prioritize TFs that are involved in distal and promoter-proximal regulation using accessibility, conformation and expression data.

Modeling transcriptional regulation using gene regulatory networks based on multi-omics data sources

“…Integrating this data into the PANDA GRNs improved the prediction performance of the models when scaled relative to promoter TFBS. This improvement was also observed in the recently published extension of the TEPIC framework [ 24 ].We observed significant improvement in both cell lines despite differences in Hi-C resolution (1 Kb for GM12878 and 5 Kb for K562), however the resolution difference may account for the greater improvement in prediction for GM12878 relative to K562.…”

“…The Passing Attributes between Networks for Data Assimilation (PANDA) algorithm generates such a GRN by extracting information from heterogeneous networks built using multiple big “omics” data sources corresponding to different TF-based regulatory mechanisms [ 23 ]. Published approaches (except for a recent extension of the TEPIC framework [ 24 ]) have also not yet considered the impact of chromatin conformation on transcriptional regulation despite its increasing availability from high throughput assays such as Hi-C [ 25 ]. Condensed chromatin within the cell is heavily restructured during the process of transcription, leading to increased accessibility of gene promoters and closer physical proximity of distal transcription machinery and enhancer elements [ 26 ].…”

A hierarchical regulatory network analysis of the vitamin D induced transcriptome reveals novel regulators and complete VDR dependency in monocytes

“…However, linking regulatory regions to target genes is not easy and each method has its limitations 64 , 65 . Thus, in the present study, we combined three different approaches to predict primary vitamin D target genes from 1,25(OH) 2 D 3 -responding genes (Fig.…”

Prediction of Whole-Cell Transcriptional Response with Machine Learning

“…It quantifies the transcriptional response to a perturbation in terms of fold-change and measures its statistical significance. Data-driven prediction using machine learning of transcription to date, however, has been limited to expression level predictions from sequences or images [19][20] .These techniques are prone to generalization errors that can arise from artifacts of normalization of counts data across experiments with combinatorically large condition spaces 21 .…”