vivaxGEN-geo

vivaxGEN-geo predicts and maps the geographic origin of Plasmodium vivax infections using a 28-SNP molecular barcode, expanded SNP panels, and statistical classifiers for molecular epidemiology and surveillance.

Key Features:

• 28-SNP molecular barcode: A 28-single nucleotide polymorphism (SNP) barcode was identified to predict country of origin for P. vivax infections.
• Training dataset: The 28-SNP barcode was derived from analysis of 831 genomes from 20 countries.
• Feature selection methods: Hierarchical FST (HFST) and decision tree (DT) methodologies were used to identify informative SNPs.
• High predictive accuracy: Cross-validation evaluations report Matthews correlation coefficient (MCC) scores exceeding 0.9 across 15 countries.
• Expanded 65-SNP panel: Combination with an existing 37-SNP barcode yields a 65-SNP panel that retains MCC >0.9 across 17 countries with up to 30% missing data.
• Secondary marker identification: Several genes were identified with median MCC scores in the range 0.54–0.68 as candidate markers for low-throughput genotyping.
• Likelihood-based classifier framework: A likelihood-based classifier supports analysis in the presence of missing data and polyclonal infections.
• Surveillance marker integration: The framework supports inclusion of drug resistance and other surveillance markers alongside geographic assignment SNPs.

Scientific Applications:

• Importation mapping: Assigns probable country of origin for P. vivax infections to map imported cases and inform epidemiological investigations.
• Elimination strategy support: Provides molecular evidence to guide resource allocation and targeted interventions in malaria elimination programs.
• Surveillance of resistance markers: Enables combined monitoring of geographic origin and drug resistance or other surveillance markers.
• Field-deployable marker use: Identifies secondary gene markers suitable for rapid low-throughput genotyping in surveillance settings.

Methodology:

Analysis of 831 genomes from 20 countries using hierarchical FST (HFST) and decision tree (DT) methods identified the 28-SNP barcode; cross-validation with Matthews correlation coefficient (MCC) quantified predictive performance; a 65-SNP panel was evaluated with up to 30% missing data; and a likelihood-based classifier was implemented to handle missing data and polyclonal infections.