DanQ: a hybrid convolutional and recurrent deep neural network for quantifying the function of DNA sequences

By Daniel Quang, Xiaohui S. Xie. 2015

788 citations recorded by the source corpus.

Modeling the properties and functions of DNA sequences is an important, but challenging task in the broad field of genomics. This task is particularly difficult for noncoding DNA, the vast majority of which is still poorly understood in terms of function. A powerful predictive model for the function of noncoding DNA can have enormous benefit for both basic science and translational research because over 98% of the human genome is noncoding and 93% of disease-associated variants lie in these regions. To address this need, we propose DanQ, a novel hybrid convolutional and bi-directional long short-term memory recurrent neural network framework for predicting noncoding function de novo from sequence. In the DanQ model, the convolution layer captures regulatory motifs, while the recurrent layer captures long-term dependencies between the motifs in order to learn a regulatory “grammar” to improve predictions. DanQ improves considerably upon other models across several metrics. For some regulatory markers, DanQ can achieve over a 50% relative improvement in the area under the precision-recall curve metric compared to related models. Availability and implementation All source code is available at the github repository http://github.com/uci-cbcl/DanQ.

Subfields: Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Molecular Biology

Citations in the corpus

Latent Space Physics: Towards Learning the Temporal Evolution of Fluid Flow (2018, methodology): “While others have studied combinations of RNNs and CNNs [8, 20], we propose a hybrid architecture to decouple the prediction task from the latent space dimensions.”
Deep learning in bioinformatics: introduction, application, and perspective in big data era (2019, methodology): “As described in [198, 120], the human GRCh37 reference genome was segmented into non-overlapping 200-bp bins.”
JASPAR 2024: 20th anniversary of the open-access database of transcription factor binding profiles (2023, methodology): “Some deep learning models have improved performance when initialising their convolutional filters with PWMs derived from JASPAR profiles ( 50 ,51 ), while many models assess derived patterns by comparison with JASPAR profiles ( 47 , 48 , 52 ).”
Artificial intelligence in clinical and genomic diagnostics (2019, methodology): “When applied to genomic sequence data, AI time series algorithms appear to be especially effective at detecting functional DNA sequence elements that are indicative of gene splicing [26, 27], large-scale regulatory elements [28], and gene function [29].”
SBSM-Pro: support bio-sequence machine for proteins (2023, positive): “Hybrid predictive models combining both CNNs and RNNs are currently popular and have been applied in various computational biology domains, including DNA methylation [36], chromatin accessibility [37], and noncoding RNAs [38].”
DeepCpG: accurate prediction of single-cell DNA methylation states using deep learning (2016, methodology): “RNNs have been successfully used for modelling long-range dependencies in natural language [58, 59], acoustic signals [60] and, more recently, genomic sequences [61, 62].”
Sequence-based prediction of protein protein interaction using a deep-learning algorithm (2017, methodology): “After that, the DnaQ model constructed by Quang and coworkers achieved more than a 50% relative improvement compared to other models for predicting the function of non-coding DNA [28].”
DNABERT: pre-trained Bidirectional Encoder Representations from Transformers model for DNA-language in genome (2020, methodology): “DNABERT-TF model to predict TFBSs in the ChIP-seq enriched regions, using 690 TF ChIP-seq uniform peak profiles from ENCODE database (37) and compared with well-known and previous published TFBS prediction tools, including DeepBind (13), DeepSEA (15), Basset (14), DeepSite (46), DanQ (20) and DESSO ”
Multifaceted protein–protein interaction prediction based on Siamese residual RCNN (2019, background): “Many recent efforts adopt a Siamese encoding architecture, where encoders based on CNN (Hu et al. , 2014; Yin and Schu¨tze, 2015) and RNN (Mueller and Thyagarajan, 2016) are widely used.”
Identifying viruses from metagenomic data using deep learning (2020, background): “…biology, such as predicting protein binding speci ﬁ city using deep learning [20 – 23], predicting the effect of non-coding variants [24,25], predicting chromatin accessibility [26], calling genetic variants from millions of short reads in NGS samples [27], predicting methylation…”
Attend and Predict: Understanding Gene Regulation by Selective Attention on Chromatin (2017, background): “4 Connecting to Previous Studies In recent years, there has been an explosion of deep learning models that have led to groundbreaking performance in many fields such as computer vision [17], natural language processing [30], and computational biology [1, 27, 38, 16, 19, 29].”
FactorNet: a deep learning framework for predicting cell type specific transcription factor binding from nucleotide-resolution sequential data (2017, methodology): “…(PWMs), which are popular models for describing the sequence-specific binding pattern of TFs. Examples of genomic application of CNNs include DanQ(Quang and Xie, 2016), DeepSEA (Zhou and Troyanskaya, 2015), Basset (Kelley et al., 2016), DeepBind (Alipanahi et al., 2015), and DeeperBind…”
An equivariant Bayesian convolutional network predicts recombination hotspots and accurately resolves binding motifs (2018, background): “Deep learning approaches have been shown to be very effective in building sequence models on large-scale data (Kelley et al., 2016; Quang and Xie, 2016; Zhou and Troyanskaya, 2015).”
NCNet: Deep Learning Network Models for Predicting Function of Non-coding DNA (2019, methodology): “(Quang and Xie, 2016) proposed a DanQ model, similar to DeepSEA, but as a hybrid framework integrating a CNN and bi-directional LSTM RNN for predicting the noncoding function de novo from a sequence.”
Deep Motif Dashboard: Visualizing and Understanding Genomic Sequences Using Deep Neural Networks (2016, methodology): “Quang and Xie [19] use the same visualization method on their convolutional-recurrent model for noncoding variant prediction.”
DeepEnhancer: Predicting enhancers by convolutional neural networks (2016, background): “Recent studies have shown that convolutional neural networks allow direct training on the DNA sequences successfully, without the need to predefine any feature sets [14]–[16], [18].”