Abstract
BACKGROUND: The discovery of RNA 5-methyluridine (m5U) modifications is vital in computational biology due to their essential significance in different biological processes. This study presents a powerful predictor named 5-meth-Uri, which improves the overall accuracy of m5U modification predictions. The proposed method combines a composite of dinucleotide and trinucleotide-based auto-cross covariance with six physicochemical parameters to generate a feature vector. We select essential features using an unsupervised Principal Component Analysis (PCA) technique to enhance the model's efficiency. Finally, an intelligent computation Deep Neural Network (DNN) was utilized to classify RNA 5-methyluridine (m5U). RESULTS: The model's performance was evaluated on two benchmark datasets, namely mature mRNA and full transcript, using tenfold cross-validation and an independent test. The proposed model achieved an average accuracy of 95.13% and 97.36% on the training set and 95.73% and 96.51% on the independent test sets for the full transcript and mature mRNA datasets, respectively. 5-meth-Uri demonstrated around 7.15% and 3.98% higher accuracy on training data and 5.42% and 3.52% higher accuracy on independent samples for both datasets compared to existing models. CONCLUSIONS: Its high accuracy and reliability make 5-meth-Uri a promising tool for researchers with potential biomedical and pharmaceutical applications.