Discovering differential genome sequence activity with interpretable and efficient deep learning

Hammelman, Jennifer; Gifford, David K.

doi:10.1371/journal.pcbi.1009282

Cited by 10 publications

(5 citation statements)

References 41 publications

(57 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To identify putative regulators of maturation, we performed motifenrichment analysis on accessible regions using two distinct but complementary approaches. We used a newly developed convolutional neural network-based method called DeepAccess 65,66 which learns the relationship between primary DNA sequence and chromatin accessibility at different ages and then predicts transcription factor motifs that drive differential chromatin accessibility at one age compared to another. In addition, we used HOMER, a de novo motif discovery method that identifies enriched motifs in provided sets of age-specific accessible peaks 67 (Methods).…”

Section: Motif Analysis Identifies Putative Regulators Of Maturationmentioning

confidence: 99%

“…Chromosomes 18 and 19 are held out for validation and testing. Methods for computing differential expected pattern effects between cell types are described in 65 . Briefly, we compute a differential expected pattern effect as the ratio between the effect that the presence of a transcription factor motif has on the predicted accessibility of a DNA sequence in one cell type relative to another cell type within a DeepAccess model.…”

Section: Non-negative Least Squares Deconvolution Of Bulk P56 Datamentioning

confidence: 99%

See 1 more Smart Citation

Transcriptional dynamics of murine motor neuron maturation in vivo and in vitro

Patel

Hammelman²,

Aziz

et al. 2022

Nat Commun

Self Cite

View full text Add to dashboard Cite

Neurons born in the embryo can undergo a protracted period of maturation lasting well into postnatal life. How gene expression changes are regulated during maturation and whether they can be recapitulated in cultured neurons remains poorly understood. Here, we show that mouse motor neurons exhibit pervasive changes in gene expression and accessibility of associated regulatory regions from embryonic till juvenile age. While motifs of selector transcription factors, ISL1 and LHX3, are enriched in nascent regulatory regions, motifs of NFI factors, activity-dependent factors, and hormone receptors become more prominent in maturation-dependent enhancers. Notably, stem cell-derived motor neurons recapitulate ~40% of the maturation expression program in vitro, with neural activity playing only a modest role as a late-stage modulator. Thus, the genetic maturation program consists of a core hardwired subprogram that is correctly executed in vitro and an extrinsically-controlled subprogram that is dependent on the in vivo context of the maturing organism.

show abstract

Section: Motif Analysis Identifies Putative Regulators Of Maturationmentioning

confidence: 99%

Section: Non-negative Least Squares Deconvolution Of Bulk P56 Datamentioning

confidence: 99%

Transcriptional dynamics of murine motor neuron maturation in vivo and in vitro

Patel

Hammelman²,

Aziz

et al. 2022

Nat Commun

Self Cite

View full text Add to dashboard Cite

show abstract

“…Hammelman and Gifford created such a deep learning approach and used it for the identification of cell state‐specific TFs in chromatin accessibility data [79, 80]. Their method is part of the framework DeepAccess and is based on an ensemble of convolutional neural networks.…”

Section: Computational Tools For Tfa Inferencementioning

confidence: 99%

“…Hammelman and Gifford created such a deep learning approach and used it for the identification of cell state-specific TFs in chromatin accessibility data [79,80]. Their method is part of the frame- Worth mentioning is also BPNet which does not predict accessibility, but the binding profile of specific TFs from CHIP-exo data [82].…”

Section: Sequence-based Deep Learning Modelsmentioning

confidence: 99%

Computational tools for inferring transcription factor activity

Hecker,

Lauber,

Behjati Ardakani

et al. 2023

Proteomics

View full text Add to dashboard Cite

Transcription factors (TFs) are essential players in orchestrating the regulatory landscape in cells. Still, their exact modes of action and dependencies on other regulatory aspects remain elusive. Since TFs act cell type‐specific and each TF has its own characteristics, untangling their regulatory interactions from an experimental point of view is laborious and convoluted. Thus, there is an ongoing development of computational tools that estimate transcription factor activity (TFA) from a variety of data modalities, either based on a mapping of TFs to their putative target genes or in a genome‐wide, gene‐unspecific fashion. These tools can help to gain insights into TF regulation and to prioritize candidates for experimental validation. We want to give an overview of available computational tools that estimate TFA, illustrate examples of their application, debate common result validation strategies, and discuss assumptions and concomitant limitations.

show abstract

“…Conversely, prevailing post hoc model interpretability methods concentrate primarily on the analysis of motifs [12][13][14][15][16][17][18][19][20][21][22] , short DNA sequences associated with regulatory functions. As sequence inputs for DNNs grow longer, deciphering the complex coordination of motifs at a scale of hundreds of kilobases (kb) becomes increasingly difficult to interpret.…”

Section: Introductionmentioning

confidence: 99%

InterpretingCis-Regulatory Interactions from Large-Scale Deep Neural Networks for Genomics

Toneyan¹,

Koo²

2023

Preprint

View full text Add to dashboard Cite

The rise of large-scale, sequence-based deep neural networks (DNNs) for predicting gene expression has introduced challenges in their evaluation and interpretation. Current evaluations align DNN predictions with experimental perturbation assays, offering a limited perspective of the DNN's capabilities within the studied loci. Moreover, existing model explainability tools mainly focus on motif analysis, which becomes complex to interpret for longer sequences. Here we introduce CREME, an in silico perturbation toolkit that interrogates large-scale DNNs to uncover rules of gene regulation that it has learned. Using CREME, we investigate Enformer, a prominent DNN in gene expression prediction, revealing cis-regulatory elements (CREs) that directly enhance or silence target genes. We explore the relationship between CRE distance from transcription start sites and gene expression, as well as the intricate complexity of higher-order CRE interactions. This work advances the ability to translate the powerful predictions of large-scale DNNs to study open questions in gene regulation.

show abstract

Discovering differential genome sequence activity with interpretable and efficient deep learning

Cited by 10 publications

References 41 publications

Transcriptional dynamics of murine motor neuron maturation in vivo and in vitro

Transcriptional dynamics of murine motor neuron maturation in vivo and in vitro

Computational tools for inferring transcription factor activity

InterpretingCis-Regulatory Interactions from Large-Scale Deep Neural Networks for Genomics

Contact Info

Product

Resources

About