A machine learning framework for promoter identification: integrating explainable AI for genomic insights

Abstract Identification of promoter regions in genomic sequences is essential for under- standing gene regulation and cellular processes. Traditional methods, often based on heuristics, face significant limitations in terms of accuracy and scalability. This study presents a machine learning-based framework for classifying promoter and non-promoter sequences by extracting and analyzing key genomic features such as GC content, CpG ratio, k-mer frequencies, and DNA bendability. We evaluated a variety of classifiers, including Logistic Regression, Support Vector Machines (SVM), Neural Networks, and XGBoost, and applied Explainable AI (XAI) techniques, such as LIME, to enhance the interpretability and transparency of the model predictions. Our results show that our proposed model outperforms traditional rule-based methods, achieving classification accuracies of up to 99.3%, with XAI providing insights into the biological relevance of the key genomic features. This work thus offers a scalable, accurate, and interpretable solution for promoter region identification, with implications for advancing genomic research and personalized medicine.