ScholarWorks@성균관대학교

초록

Accurate identification of enhancer regions in DNA sequences is essential for understanding gene regulation and its role in diverse biological processes. Enhancers are regulatory elements that influence gene expression, but their detection remains challenging due to the complexity and variability of genomic sequences. In this study, we propose AttnW2V-Enhancer, a novel model that combines Word2Vec-based sequence encoding, convolutional neural networks (CNN), and attention mechanisms to address this challenge. By leveraging Word2Vec embeddings, our model captures biologically meaningful patterns and offers a more efficient and interpretable representation than traditional methods such as one-hot encoding and physicochemical descriptors. We evaluate AttnW2V-Enhancer on an independent test set, where it achieves superior performance with an accuracy of 81.75%, sensitivity of 83.50%, specificity of 80.00%, and a Matthews Correlation Coefficient (MCC) of 0.635, outperforming existing models. Additionally, we demonstrate the effectiveness of the attention mechanism in enhancing feature learning by dynamically focusing on the most relevant sequence regions. These results confirm that integrating Word2Vec encoding with CNNs and attention mechanisms provides a powerful and interpretable framework for enhancer prediction, offering valuable insights into the identification of regulatory sequences. The source code and implementation are publicly available at: https://github.com/Rehman1995/AttnW2V-Enhancer. © 2025

키워드