Enhancer-WVM: Prediction of Enhancers and Their Strengths Based on Integrated Learning

Enhancers are an important class of noncoding DNA elements in the genome that can precisely regulate gene expression by binding to transcription factors, thus playing a key role in cell differentiation, development and stress response. However, in the face of the timeconsuming and costly nature of traditional biomedical assays, there is an urgent need for a deep learning-based enhancer prediction model. In this paper, we propose an integrated learning-based enhancer prediction model Enhancer-WVM for predicting enhancers and their strengths in DNA sequences. We designed a convolutional neural network (CNN), a bidirectional temporal neural network (BiLSTM), and a hybrid model combining multi-scale convolution and attention mechanisms, and model integration was achieved through 10 -fold cross-validation and weighted soft voting. Sequence features are generated using one-hot (One-hot) coding and nucleotide chemical property (NCP) coding, and the training data are augmented by random shifting and backward complementation. Experimental results show that the framework achieves excellent performance in the augmented subclassification task, with sensitivity (Sn), specificity (Sp), accuracy (Acc), Matthews correlation coefficient (MCC), and area under the ROC curve (AUC) of 81 %, 75.75 %, 78.37 %, 56.83 %, and 86.99 %, respectively. This study demonstrates a robust and efficient computational method that reduces training costs and improves prediction accuracy.