Hepatitis C Virus Genotyping with a Transformer Neural Network

Summary

This study evaluated the performance of Transformer-based models in predicting hepatitis C virus (HCV) genotypes and subtypes after fine-tuning. A total of 2,881 HCV whole-genome sequences from the Los Alamos HCV Sequence Database were used, covering genotypes 1 to 6 and all confirmed subtypes; genotypes 7 and 8 were excluded due to insufficient samples. Fine-tuning with homogeneous fragmentation and balanced sample distribution achieved precision ranging from 98.48% to 100% for genotype classification. In contrast, using a fragmentation strategy that caused data imbalance and arbitrary sample distribution across training files resulted in a precision of 48.12%, which is considered low, with a high error rate for genotype 5 due to its low frequency in the datasets. For subtype classification, the best-performing fine-tuning approach achieved 99.89% accuracy and 99.87% precision. Models that included additional genotypes showed a slight performance decrease due to increased task complexity. The study demonstrates that, with properly fragmented, distributed, and labeled data, Transformer-based neural networks are a promising approach for HCV genotype and subtype classification.