Exploring NS5A Protein of Hepatitis C Virus Through Computational Deep Learning Models and Phylogenetic Analysis for Sequence Variability

Abstract

Background: Hepatitis C Virus (HCV) remains a global health threat, causing severe liver diseases such as cirrhosis and hepatocellular carcinoma. Protein sequences are essential for exploring genetic variability and improving detection. This study presents a generative LSTM model, AI-VNA, for generating and analyzing NS5A protein sequences to enhance understanding of sequence diversity and improve predictive modeling. Methods: The AI-VNA model was trained on curated NS5A protein sequences and evaluated against baseline models including CNN, 1D-CNN, GRU, and Simple RNN. Generated sequences were validated using BLAST (version 2.16.0) to assess biological relevance. Additional analyses included a sequence similarity heatmap, phylogenetic clustering using UPGMA, and statistical comparison of sequence similarities via the Mann–Whitney U test. Results: AI-VNA achieved 98.56% accuracy, outperforming other models and demonstrating strong capability in modeling sequence patterns. BLAST validation showed an average query coverage of 82%, an average E-value of 1 × 10−155, and 96.68% identity to known NS5A proteins. The sequence similarity heatmap revealed notable variability among the generated sequences. Phylogenetic analysis showed distinct clustering patterns, indicating both conservation and diversity. A box plot comparing similarity scores of predicted and database sequences showed significant variation (p = 0.01221; Cliff’s Delta = 0.291, small effect). Conclusions: These findings highlight the AI-VNA model’s ability to generate evolutionarily relevant and diverse NS5A protein sequences, supporting its potential for improving HCV sequence analysis and aiding in diagnostic and therapeutic developments.

The authors' abstract, as published at the source. Diagnostics, 2026 · DOI ↗