In silico prediction of high-resolution Hi-C interaction matrices

Zhang, Shilu; Chasman, Deborah; Knaack, Sara; Roy, Sushmita

doi:10.1038/s41467-019-13423-8

Cited by 63 publications

(68 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We show that gradient boosting achieve the best predictions, yielding accuracies of ~ 95%. Moreover, we observe that chromatin features close to anchor regions contribute most to the predictions, in line with previous observations [15–17]. We also find that gradient boosting models trained with transcription-associated signals alone, unlike the other algorithms tested, are able to generate accurate predictions in K562 cell line, with transcription factors at the anchors being among the most predictive features.…”

Section: Resultssupporting

confidence: 91%

“…Therefore, in silico predictions that take advantage of the wealth of publicly available sequencing data emerges as a rational strategy to generate virtual chromatin interaction maps in new cell types for which experimental maps are still lacking. To date, several studies have been devoted to predict chromatin loops based on one-dimensional (1D) genomic information with accurate results [13–17]. In such works, authors have modeled loops using different designs and machine learning approaches.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A comparative study of supervised machine learning algorithms for the prediction of long-range chromatin interactions

Vanhaeren

Divina

García-Torres

et al. 2020

Preprint

View full text Add to dashboard Cite

The role of three-dimensional genome organization as a critical regulator of gene expression 1 has become increasingly clear over the last decade. Most of our understanding of this association 2 comes from the study of long range chromatin interaction maps provided by Chromatin Conformation 3 Capture-based techniques, which have greatly improved in recent years. Since these procedures are 4 experimentally laborious and expensive, in silico prediction has emerged as an alternative strategy 5 to generate virtual maps in cell types and conditions for which experimental data of chromatin 6 interactions is not available. Several methods have been based on predictive models trained on 7 one-dimensional (1D) sequencing features, yielding promising results. However, different approaches 8 vary both in the way they model chromatin interactions and in the machine learning-based strategy 9 they rely on, making it challenging to carry out performance comparison of existing methods. In this 10 study, we use publicly available 1D sequencing signals to model chromatin interactions in two human 11 cell lines and evaluate the prediction performance of 5 popular machine learning algorithms: decision 12 trees, random forests, gradient boosting, support vector machines and multi-layer perceptron. Our 13 approach accurately predicts long-range interactions and reveals that gradient boosting significantly 14 outperforms the other four algorithms, yielding accuracies of ∼ 95%. We show that chromatin features 15 in close genomic proximity to the anchors cover most of the predictive information. Moreover, we 16 demonstrate that gradient boosting models trained with different subsets of chromatin features, 17 unlike the other methods tested, are able to produce accurate predictions. In this regard, and 18 besides architectural proteins, transcription factors are shown to be highly informative. Our study 19 provides a framework for the systematic prediction of long-range chromatin interactions, identifies 20 gradient boosting as the best suited algorithm for this task and highlights cell-type specific binding 21 of transcription factors at the anchors as important determinants of chromatin wiring. 22

show abstract

Section: Resultssupporting

confidence: 91%

Section: Introductionmentioning

confidence: 99%

A comparative study of supervised machine learning algorithms for the prediction of long-range chromatin interactions

Vanhaeren

Divina

García-Torres

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…from replicates, without accounting for the distance dependence signal leads to inflated correlation values 40 . To mitigate this effect, it is possible to stratify the correlation over the distance between interacting regions 40,41 . The Hi-C skeleton transformation employed already accounts for the distance dependence in terms of DNA polymer proximity.…”

Section: Distance Stratified Correlationmentioning

confidence: 99%

DeepC: Predicting chromatin interactions using megabase scaled deep neural networks and transfer learning

Gosden

Downes

Brown

et al. 2019

Preprint

View full text Add to dashboard Cite

Understanding 3D genome structure requires high throughput, genome-wide approaches. However, assays for all vs. all chromatin interaction mapping are expensive and time consuming, which severely restricts their usage for large-scale mutagenesis screens or for mapping the impact of sequence variants. Computational models sophisticated enough to grasp the determinants of chromatin folding provide a unique window into the functional determinants of 3D genome structure as well as the effects of genome variation. A chromatin interaction predictor should work at the base pair level but also incorporate large-scale genomic context to simultaneously capture the large scale and intricate structures of chromatin architecture. Similarly, to be a flexible and generalisable approach it should also be applicable to data it has not been explicitly trained on. To develop a model with these properties, we designed a deep neuronal network (deepC) that utilizes transfer learning to accurately predict chromatin interactions from DNA sequence at megabase scale. The model generalizes well to unseen chromosomes and works across cell types, Hi-C data resolutions and a range of sequencing depths. DeepC integrates DNA sequence context on an unprecedented scale, bridging the different levels of resolution from base pairs to TADs. We demonstrate how this model allows us to investigate sequence determinants of chromatin folding at genome-wide scale and to predict the importance of regulatory elements and the impact of sequence variations.

show abstract

“…In addition, we chose to train MATCHA using multi-way interaction data only in this work, but the model can easily include other functional genomic signals as features of the corresponding genomic bins. This would in principle extend the existing work on predicting pairwise chromatin loops based on functional genomic data (Huang et al, 2015;Whalen et al, 2016;Zhang et al, 2018;Kai et al, 2018;Zhang et al, 2019). Finally, the multi-way chromatin interactions extracted by MATCHA could also be the foundation to connect transcriptional regulation and 3D genome organization (Stadhouders et al, 2019;Kim and Shendure, 2019;Tian et al, 2020).…”

Section: Discussionmentioning

confidence: 85%

Probing multi-way chromatin interaction with hypergraph representation learning

Zhang

2020

Preprint

View full text Add to dashboard Cite

Advances in high-throughput mapping of 3D genome organization have enabled genome-wide characterization of chromatin interactions. However, proximity ligation based mapping approaches for pairwise chromatin interaction such as Hi-C cannot capture multi-way interactions, which are informative to delineate higher-order genome organization and gene regulation mechanisms at singlenucleus resolution. The very recent development of ligation-free chromatin interaction mapping methods such as SPRITE and ChIA-Drop has offered new opportunities to uncover simultaneous interactions involving multiple genomic loci within the same nuclei. Unfortunately, methods for analyzing multi-way chromatin interaction data are significantly underexplored. Here we develop a new computational method, called MATCHA, based on hypergraph representation learning where multiway chromatin interactions are represented as hyperedges. Applications to SPRITE and ChIA-Drop data suggest that MATCHA is effective to denoise the data and make de novo predictions of multiway chromatin interactions, reducing the potential false positives and false negatives from the original data. We also show that MATCHA is able to distinguish between multi-way interaction in a single nucleus and combination of pairwise interactions in a cell population. In addition, the embeddings from MATCHA reflect 3D genome spatial localization and function. MATCHA provides a promising framework to significantly improve the analysis of multi-way chromatin interaction data and has the potential to offer unique insights into higher-order chromosome organization and function.

show abstract

In silico prediction of high-resolution Hi-C interaction matrices

Cited by 63 publications

References 53 publications

A comparative study of supervised machine learning algorithms for the prediction of long-range chromatin interactions

A comparative study of supervised machine learning algorithms for the prediction of long-range chromatin interactions

DeepC: Predicting chromatin interactions using megabase scaled deep neural networks and transfer learning

Probing multi-way chromatin interaction with hypergraph representation learning

Contact Info

Product

Resources

About