Penta‑Class Classification of Hepatitis Virus DNA Sequences Using a 1D‑CNN for Enhanced Differential Diagnosis

Abstract

Despite advances in molecular testing, the genetic similarity among hepatitis strains and the complexity of their clinical presentations often confound traditional diagnostic approaches, underscoring the urgent need for automated, sequence‑intelligent solutions. This study developed a pure one‑dimensional Convolutional Neural Network (1D‑CNN) to classify five hepatitis virus types (HAV, HBV, HCV, HDV, HEV) directly from raw DNA sequences, eliminating complex preprocessing such as k‑mer segmentation or external optimization. A total of 500 complete genomic sequences (100 per class) were retrieved from the NCBI Virus database. Following one‑hot encoding and sequence padding, the proposed 1D‑CNN architecture—employing convolutional feature extraction with batch normalization and regularization—was trained using the Adam optimizer and categorical cross‑entropy loss. Performance was evaluated using accuracy, precision, recall, F1‑score, Matthews Correlation Coefficient (MCC), ROC‑AUC, and precision‑recall curves. The model achieved an overall accuracy of 95%, MCC of 0.9395, macro precision of 0.9571, and macro recall of 0.9500. ROC‑AUC values reached 1.00 for four classes and 0.97 for HDV, while precision‑recall average precision ranged from 0.971 to 1.00. The confusion matrix revealed minimal misclassifications, primarily between HEV and HDV, confirming that the model autonomously extracts discriminative nucleotide patterns for reliable multi‑species classification. This study contributes an alignment‑free, computationally efficient, and reproducible approach for hepatitis virus typing, outperforming many previous methods reliant on manual feature engineering or heuristic optimization. Future work should validate the model on clinical samples and explore the interpretability of learned motifs.