Identification of recent cases of hepatitis C virus infection using physical-chemical properties of hypervariable region 1 and a radial basis function neural network classifier

Abstract

RESULTS: Using dinucleotide auto correlation (DAC), we identified physical-chemical (PhyChem) features of HVR1 variants. Significant (p < 9.58 × 10-4) differences in the means and frequency distributions of PhyChem features were found between HVR1 variants sampled from patients with recent vs chronic (R/C) infection. Moreover, the R-associated variants were found to occupy distinct and discrete PhyChem spaces. A radial basis function neural network classifier trained on the PhyChem features of intra-host HVR1 variants accurately classified R/C-HVR1 variants (classification accuracy (CA) = 94.85%; area under the ROC curve, AUROC = 0.979), in 10-fold cross-validation). The classifier was accurate in assigning individual HVR1 variants to R/C-classes in the testing set (CA = 84.15%; AUROC = 0.912) and in detection of infection duration (R/C-class) in patients (CA = 88.45%). Statistical tests and evaluation of the classifier on randomly-labeled datasets indicate that classifiers' CA is robust (p < 0.001) and unlikely due to random correlations (CA = 59.04% and AUROC = 0.50). CONCLUSIONS: The PhyChem features of intra-host HVR1 variants are strongly associated with the duration of HCV infection. Application of the PhyChem biomarkers to models for detection of the R/C-state of HCV infection in patients offers a new opportunity for detection of outbreaks and for molecular surveillance. The method will be available at https://webappx.cdc.gov/GHOST/ to the authenticated users of Global Hepatitis Outbreak and Surveillance Technology (GHOST) for further testing and validation. BACKGROUND: Identification of acute or recent hepatitis C virus (HCV) infections is important for detecting outbreaks and devising timely public health interventions for interruption of transmission. Epidemiological investigations and chemistry-based laboratory tests are 2 main approaches that are available for identification of acute HCV infection. However, owing to complexity, both approaches are not efficient. Here, we describe a new sequence alignment-free method to discriminate between recent (R) and chronic (C) HCV infection using next-generation sequencing (NGS) data derived from the HCV hypervariable region 1 (HVR1).