Multinomial Convolutions for Joint Modeling of Sequence Motifs and Enhancer Activities

Park, Minjun; Singh, Salvi; Canozo, Francisco Jose Grisanti; Samee, Md. Abul Hassan

doi:10.1101/2020.07.28.224212

Cited by 4 publications

(6 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The rst two layers are applied parallelly to both the forward and the reverse complement complement strand of the input sequence, allowing the model to detect patterns in both DNA strands. The rst layer is a multinomial convolution layer with ReLU activation 10 . Filters in this layer detect motifs and the convolution operation computes matching scores of the motifs (log likelihood ratios, see Methods) at each position of the input sequence.…”

Section: Deepbend: a Deep Convolutional Neural Network (Cnn) Model Of...mentioning

confidence: 99%

DeepBend: An interpretable model of DNA bendability

Khan¹,

Sakib²,

Rahman³

et al. 2023

iScience

Self Cite

View full text Add to dashboard Cite

Section: Deepbend: a Deep Convolutional Neural Network (Cnn) Model Of...mentioning

confidence: 99%

DeepBend: An interpretable model of DNA bendability

Khan¹,

Sakib²,

Rahman³

et al. 2023

iScience

Self Cite

View full text Add to dashboard Cite

“…Interpreting the DeepBend model is straight-forward. Since each row of a first layer filter is a multinomial distribution over the four nucleotides, these filters are directly interpretable as biophysical models of sequence motifs 10 . Regularising variance in the last layer separates out the relative spatial patterns of motifs significant for different ranges of bendability into different filters of the second convolution layer.…”

Section: Resultsmentioning

confidence: 99%

“…The first layer is the custom 1D convolution from the MuSeAM model 10 that learns the multinomial distribution of motif patterns. In this multinomial convolutional layer, 256 filters of size 8×4 have been used with “same” padding and stride 1.…”

Section: Methodsmentioning

confidence: 99%

“…The first two layers are applied parallelly to both the forward and the reverse complement complement strand of the input sequence, allowing the model to detect patterns in both DNA strands. The first layer is a multinomial convolution layer with ReLU activation 10 . Filters in this layer detect motifs and the convolution operation computes matching scores of the motifs (log likelihood ratios, see Methods) at each position of the input sequence.…”

Section: Deepbend: a Deep Convolutional Neural Network (Cnn) Model Of...mentioning

confidence: 99%

See 1 more Smart Citation

DeepBend: An Interpretable Model of DNA Bendability

Khan

Sakib

Rahman

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

The bendability of genomic DNA impacts chromatin packaging and protein-DNA binding. However, beyond a handful of known sequence motifs, such as certain dinucleotides and poly(A)/poly(T) sequences, we do not have a comprehensive understanding of the motifs influencing DNA bendability. Recent high-throughput technologies like Loop-Seq offer an opportunity to address this gap but the lack of accurate and interpretable machine learning models still poses a significant challenge. Here we introduce DeepBend, a convolutional neural network model built as a visible neural network where we designed the convolutions to directly capture the motifs underlying DNA bendability and how their periodic occurrences or relative arrangements modulate bendability. Through extensive benchmarking on Loop-Seq data, we show that DeepBend consistently performs on par with alternative machine learning and deep learning models while giving an extra edge through mechanistic interpretations. Besides confirming the known motifs of DNA bendability, DeepBend also revealed several novel motifs and showed how the spatial patterns of motif occurrences influence bendability. DeepBend's genome-wide prediction of bendability further showed how bendability is linked to chromatin conformation and revealed the motifs controlling bendability of topologically associated domains and their boundaries.

show abstract

“…In order to build CNN models that are both accurate and easy-to-interpret, we used the recently proposed MuSeAM model (Fig 1A). MuSeAM is a custom CNN model shown to be more interpretable than conventional CNN for sequence analysis [20]. We used the model for the prediction of DNA binding specificity of EBNA2 and EBNA3 vTFs.…”

mentioning

confidence: 99%

Epstein-Barr virus nuclear antigen proteins deploy diverse mechanisms to bind the human genome

Haque

Tushar

Park

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Viral transcription factors (vTFs) are known to bind the human genome and impact critical biological processes. However, a vTF can infect multiple cell-types, and the extent to which they deploy the same mechanisms for DNA binding across different cell-types is poorly understood. We used the Epstein-Barr virus Nuclear Antigen Proteins 2 and 3 (EBNA2 and EBNA3) to address this gap, the two widely studied vTFs associated with cancers, as case studies. We analyze multiple ChIP-seq datasets of these vTFs from different human cell-lines using a state-of-the-art convolutional neural network model. We show that each vTF uses both cell-type specific and cell-type independent cofactors to bind the human genome. Interestingly, each vTF requires ∼10 cofactors consistently across all binding peaks. Such dependency on a large number of cofactors is in contrast to human TFs since human TFs typically co-bind with 2-3 cofactors. Our de novo motif finding approach also reveals novel sequence motifs at the ChIP peaks of these vTFs. These critical signals were missed by previous studies doing enrichment analysis using known human TFs. Finally, we find that although a vTF impacts similar biological processes across cell-types, it also impacts distinct biological processes in a cell-type specific manner. Our study provides a roadmap for integrative analysis of vTF binding and pinpointing their diverse mechanisms of targeting the human genome across different human cell-types.

show abstract

Multinomial Convolutions for Joint Modeling of Sequence Motifs and Enhancer Activities

Cited by 4 publications

References 59 publications

DeepBend: An interpretable model of DNA bendability

DeepBend: An interpretable model of DNA bendability

DeepBend: An Interpretable Model of DNA Bendability

Epstein-Barr virus nuclear antigen proteins deploy diverse mechanisms to bind the human genome

Contact Info

Product

Resources

About