DeepDualEnhancer: A Dual-Feature Input DNABert Based Deep Learning Method for Enhancer Recognition

Abstract

Enhancers are cis-regulatory DNA sequences that are widely distributed throughout the genome. They can precisely regulate the expression of target genes. Since the features of enhancer segments are difficult to detect, we propose DeepDualEnhancer, a DNABert-based method using a multi-scale convolutional neural network, BiLSTM, for enhancer identification. We first designed the DeepDualEnhancer method based only on the DNA sequence input. It mainly consists of a multi-scale Convolutional Neural Network, and BiLSTM to extract features by DNABert and embedding, respectively. Meanwhile, we collected new datasets from the enhancer-promoter interaction field and designed the method DeepDualEnhancer-genomic for inputting DNA sequences and genomic signals, which consists of the transformer sequence attention. Extensive comparisons of our method with 20 other excellent methods through 5-fold cross validation, ablation experiments, and an independent test demonstrated that DeepDualEnhancer achieves the best performance. It is also found that the inclusion of genomic signals helps the enhancer recognition task to be performed better.

Keywords: DNABert; deep learning; enhancer; genomic signal.

Figure 1

DeepDualEnhancer motifs analyzed by MEME suite.

Figure 2

Comparison of DeepDualEnhancer with four selected well-performing methods (Stage 1).

Figure 3

Comparison of DeepDualEnhancer with four selected well-performing methods (Stage 2).

Figure 4

Results of ablation experiments. (a) and (b) indicate the comparison of experimental results at Stage 1 and Stage 2, respectively.

Figure 5

Comparison results on new datasets. (a) ACC metrics. (b) MCC metrics. (c) AUC metrics.

Figure 5

Comparison results on new datasets. (a) ACC metrics. (b) MCC metrics. (c) AUC metrics.

Figure 6

t-SNE classification results for embedding (left) and DNABert (right). Here, 0.0 and 1.0 represent label classification, and the x-coordinate and y-coordinate represent the distance after dimensionality reduction.

Figure 7

Results of genomic characterization ablation experiments on balanced BENGI dataset: (a) HMEC cell line; (b) IMR90 cell line; (c) K562 cell line; (d) NHEK cell line.

Figure 8

The percentage of different chromosomes on the BENGI Dataset.

Figure 9

The results of the ROC curve for the independent chromosome test set.

Figure 10

A snapshot of the dataset. (a) A sample of the input DNA sequence. (b) Introduction to genomic signals used in the new dataset. (c) Base distribution in the dataset. (d) The distribution of samples on different chromosomes in the new dataset. (e) Distribution of CTCF signal frequency on six cell lines. (f) The flowchart of this study.

Figure 11

Network architecture of DeepDualEnhancer.

Figure 12

The architecture of DNABert.

Figure 13

Network architecture of DeepDualEnhancer-genomic.