BindVAE: Dirichlet variational autoencoders for de novo motif discovery from accessible chromatin

Kshirsagar, Meghana; Yuan, Han; Ferres, Juan Lavista; Leslie, Christina

doi:10.1186/s13059-022-02723-w

Cited by 8 publications

(7 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To the best of our knowledge, just four other tools exist intended to generate de novo motifs from ATAC-seq data, namely BindVAE 15 and MMGraph 16 (both using machine learning in combination with k-mers), CEMIG 17 (which utilizes De Bruijin graphs created on k-mers), and the RSAT peak-motifs pipeline 18 (a pipeline intended for ChIP-seq, which is also applicable to ATAC-seq data). However, BindVAE, CEMIG and RSAT solely operate on the sequences of complete ATAC-seq peaks, therefore these tools lack precision compared to a FP based tool.…”

Section: Resultsmentioning

confidence: 99%

Uncovering uncharacterized binding of transcription factors from ATAC-seq footprinting data

Schultheis,

Bentsen,

Heger

et al. 2024

Sci Rep

View full text Add to dashboard Cite

Transcription factors (TFs) are crucial epigenetic regulators, which enable cells to dynamically adjust gene expression in response to environmental signals. Computational procedures like digital genomic footprinting on chromatin accessibility assays such as ATACseq can be used to identify bound TFs in a genome-wide scale. This method utilizes short regions of low accessibility signals due to steric hindrance of DNA bound proteins, called footprints (FPs), which are combined with motif databases for TF identification. However, while over 1600 TFs have been described in the human genome, only ~ 700 of these have a known binding motif. Thus, a substantial number of FPs without overlap to a known DNA motif are normally discarded from FP analysis. In addition, the FP method is restricted to organisms with a substantial number of known TF motifs. Here we present DENIS (DE Novo motIf diScovery), a framework to generate and systematically investigate the potential of de novo TF motif discovery from FPs. DENIS includes functionality (1) to isolate FPs without binding motifs, (2) to perform de novo motif generation and (3) to characterize novel motifs. Here, we show that the framework rediscovers artificially removed TF motifs, quantifies de novo motif usage during an early embryonic development example dataset, and is able to analyze and uncover TF activity in organisms lacking canonical motifs. The latter task is exemplified by an investigation of a scATAC-seq dataset in zebrafish which covers different cell types during hematopoiesis.

show abstract

Section: Resultsmentioning

confidence: 99%

Uncovering uncharacterized binding of transcription factors from ATAC-seq footprinting data

Schultheis,

Bentsen,

Heger

et al. 2024

Sci Rep

View full text Add to dashboard Cite

show abstract

“…To the best of our knowledge, just four other tools exist intended to generate de novo motifs from ATAC-seq data, namely BindVAE 11 and MMGraph 12 (both using machine learning in combination with k-mers), CEMIG 13 (which utilizes De Bruijin graphs created on k-mers), and the RSAT peak-motifs pipeline 14 (a pipeline intended for ChIP-seq, which is also applicable to ATAC-seq data). However, BindVAE, CEMIG and RSAT solely operate on the sequences of complete ATAC-seq peaks, therefore these tools lack precision compared to a FP based tool.…”

Section: Resultsmentioning

confidence: 99%

Uncovering uncharacterized binding of transcription factors from ATAC-seq footprinting data

Schultheis,

Bentsen,

Heger

et al. 2023

Preprint

View full text Add to dashboard Cite

Transcription factors (TFs) are crucial epigenetic regulators, which enable cells to dynamically adjust gene expression in response to environmental signals. Computational procedures like digital genomic footprinting on chromatin accessibility assays such as ATACseq can be used to identify bound TFs in a genome-wide scale. This method utilizes short regions of low accessibility signals due to steric hindrance of DNA bound proteins, called footprints (FPs), which are combined with motif databases for TF identification. However, while over 1600 TFs have been described in the human genome, only ∼700 of these have a known binding motif. Thus, a substantial number of FPs without overlap to a known DNA motif are normally discarded from FP analysis. In addition, the FP method is restricted to organisms with a substantial number of known TF motifs. Here we present DENIS (DE Novo motIf diScovery), a framework to generate and systematically investigate the potential of de novo TF motif discovery from FPs. DENIS includes functionality i) to isolate FPs without binding motifs, ii) to perform de novo motif generation and iii) to characterize novel motifs. Here, we show that the framework rediscovers artificially removed TF motifs, quantifies de novo motif usage during an early embryonic development example dataset, and is able to analyze and uncover TF activity in organisms lacking canonical motifs. The latter task is exemplified by an investigation of a scATAC-seq dataset in zebrafish which covers different cell types during hematopoiesis.

show abstract

“…One shortcoming of scover is that only a minority of the motifs end up being used by the model, and this is computationally costly. By contrast, other methods such as BindVAE [ 23 ] have a more efficient motif usage model, and we believe that exploring better motif representations is an important future direction of research. Our results are also consistent with other studies that have categorized TFs as either preferentially binding promoters or enhancers [ 43 ].…”

Section: Discussionmentioning

confidence: 99%

“…One set of methods, e.g., BPNet [ 20 ], Enformer [ 21 ], and Basset [ 13 , 22 ], require a signal (e.g., chromatin accessibility) to be associated with the sequence, and they can be used to predict the values for sequences that have not been observed during training. In doing so, these methods learn sequence motifs, but with a multi-layered architecture, the representation is distributed and may be difficult to interpret biologically, although methods such as BindVAE [ 23 ] have been successful in using other approaches. This setup, however, is not ideal for predicting gene expression, so different approaches have been proposed [ 6 , 24 , 25 ].…”

Section: Introductionmentioning

confidence: 99%

Predicting the impact of sequence motifs on gene regulation using single-cell data

et al. 2023

View full text Add to dashboard Cite

The binding of transcription factors at proximal promoters and distal enhancers is central to gene regulation. Identifying regulatory motifs and quantifying their impact on expression remains challenging. Using a convolutional neural network trained on single-cell data, we infer putative regulatory motifs and cell type-specific importance. Our model, scover, explains 29% of the variance in gene expression in multiple mouse tissues. Applying scover to distal enhancers identified using scATAC-seq from the developing human brain, we identify cell type-specific motif activities in distal enhancers. Scover can identify regulatory motifs and their importance from single-cell data where all parameters and outputs are easily interpretable.

show abstract

BindVAE: Dirichlet variational autoencoders for de novo motif discovery from accessible chromatin

Cited by 8 publications

References 41 publications

Uncovering uncharacterized binding of transcription factors from ATAC-seq footprinting data

Uncovering uncharacterized binding of transcription factors from ATAC-seq footprinting data

Uncovering uncharacterized binding of transcription factors from ATAC-seq footprinting data

Predicting the impact of sequence motifs on gene regulation using single-cell data

Contact Info

Product

Resources

About