A Method for the Identification and Intensity Prediction of DNA Enhancers Based on Feature Extraction Algorithms

Abstract

DNA enhancers are important non-coding cis-regulatory elements that control gene transcription. To reduce reliance on high-cost experimental methods such as ChIP-seq and DNase-seq, this paper constructs a classification framework based on sequence features and supervised learning for two sequence-level tasks: enhancer identification (enhancer/non-enhancer) and enhancer strength prediction (strong/weak enhancer).Methodologically, k-mer frequency features and CKSNAP-based dinucleotide interval features were extracted separately, and these were concatenated to form a fused representation. Under a unified hierarchical 5-fold cross-validation framework, the performance of Support Vector Machines (SVM) and XGBoost models was compared. The results indicate that, for both tasks, XGBoost performed slightly better or was comparable to SVM under the same feature set;under the same classifier, the fusion of k-mer and CKSNAP features outperforms either feature alone. Specifically, XGBoost based on the fusion features achieved accuracy rates of 0.796 and 0.681 for enhancer identification and strength classification, respectively, with overall superior ROC/AUC performance. This indicates that local k-mer composition information and short-range distance-dependent information are complementary, thereby enhancing the model’s discriminative power and stability.