Learning representations of chromatin contacts using a recurrent neural network identifies genomic drivers of conformation

Dsouza, Kevin Bradley; Maslova, Alexandra; Al-Jibury, Ediem; Merkenschlager, Matthias; Bhargava, V.K.; Libbrecht, Maxwell W.

doi:10.1038/s41467-022-31337-w

Cited by 9 publications

(3 citation statements)

References 119 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Besides, we introduced contrastive pretraining 42 and data augmentation by downsampling Hi-C contact map techniques to train a single model capable of handling Hi-C data of different sequencing depths. We believe this training procedure can improve many machine learning applications for Hi-C data analysis 12 , 43 – 45 .…”

Section: Discussionmentioning

confidence: 99%

Reference panel guided topological structure annotation of Hi-C data

Zhang

Blanchette

2022

Nat Commun

View full text Add to dashboard Cite

Accurately annotating topological structures (e.g., loops and topologically associating domains) from Hi-C data is critical for understanding the role of 3D genome organization in gene regulation. This is a challenging task, especially at high resolution, in part due to the limited sequencing coverage of Hi-C data. Current approaches focus on the analysis of individual Hi-C data sets of interest, without taking advantage of the facts that (i) several hundred Hi-C contact maps are publicly available, and (ii) the vast majority of topological structures are conserved across multiple cell types. Here, we present RefHiC, an attention-based deep learning framework that uses a reference panel of Hi-C datasets to facilitate topological structure annotation from a given study sample. We compare RefHiC against tools that do not use reference samples and find that RefHiC outperforms other programs at both topological associating domain and loop annotation across different cell types, species, and sequencing depths.

show abstract

Section: Discussionmentioning

confidence: 99%

Reference panel guided topological structure annotation of Hi-C data

Zhang

Blanchette

2022

Nat Commun

View full text Add to dashboard Cite

show abstract

“…39,40 Even so, a standard RNN network only has the function of short-term memory and will cause problems of gradient disappearance and gradient explosion. 41 Aimed at solving these problems, an improved RNN, long short-term memory (LSTM) network is proposed, and this network exhibits greater robustness and memory capability than a standard RNN. 42 To the best of our knowledge, an LSTM network has not previously been applied to lithographic process modeling and optimization.…”

Section: Introductionmentioning

confidence: 99%

Critical dimension prediction of metal oxide nanoparticle photoresists for electron beam lithography using a recurrent neural network

Zhao

Wang

et al. 2023

Nanoscale

View full text Add to dashboard Cite

show abstract

“…Alphafold (Senior, et al, 2019) used one-dimensional (1D) ConvLSTM for predicting protein structures. Hi-C-LSTM (Dsouza, et al, 2022) used LSTM for learning low-dimensional latent representations of a Hi-C contact matrix.…”

Section: Introductionmentioning

confidence: 99%

HiC4D: Forecasting spatiotemporal Hi-C data with residual ConvLSTM

Liu

Wang

2022

Preprint

View full text Add to dashboard Cite

Motivation: The Hi-C experiments have been extensively used for the studies of mammalian genomic structures. In the last few years, spatiotemporal Hi-C has enormously contributed to the investigation of genome dynamic reorganization. However, computationally modeling and forecasting spatiotemporal Hi-C data still have not been seen in the literature. Results: We present HiC4D for dealing with the problem of forecasting spatiotemporal Hi-C data. We designed and benchmarked a novel network, which is a combination of residual network and convolutional long short-term memory (ConvLSTM), and named it residual ConvLSTM (ResConvLSTM). We evaluated our new method and compared it with the other four methods including three outstanding video-prediction methods from the literature: ConvLSTM, spatiotemporal LSTM (ST-LSTM), and simple video prediction (SimVP), and one self-designed naive network (NaiveNet) as a baseline. We used four different spatiotemporal Hi-C datasets for the blind test, including two from mouse embryogenesis, one from somatic cell nuclear transfer (SCNT) embryos, and one from human embryogenesis. Our evaluation results indicate that ResConvLSTM almost always outperforms the other four methods on the four blind-test datasets in terms of accurately reproducing spatiotemporal Hi-C contact matrices at future time steps. Our benchmarks also indicate that all five methods can successfully recover the boundaries of topologically associating domains (TADs) called on the experimental Hi-C contact matrices. Taken together, our benchmarks suggest that HiC4D is an effective, useful tool for predicting spatiotemporal Hi-C data. Availability: HiC4D is publicly available at http://dna.cs.miami.edu/HiC4D/. Contact: zheng.wang@miami.edu

show abstract

Learning representations of chromatin contacts using a recurrent neural network identifies genomic drivers of conformation

Cited by 9 publications

References 119 publications

Reference panel guided topological structure annotation of Hi-C data

Reference panel guided topological structure annotation of Hi-C data

Critical dimension prediction of metal oxide nanoparticle photoresists for electron beam lithography using a recurrent neural network

HiC4D: Forecasting spatiotemporal Hi-C data with residual ConvLSTM

Contact Info

Product

Resources

About