Genome-Wide k-Mer Signatures Identify Species of Anopheles gambiae Complex from Low-Coverage Sequencing

Species-level identification of mosquitoes within the Anopheles gambiae complex is essential for vector surveillance but remains challenging because several members are morphologically indistinguishable and share substantial genomic variation. Here, I evaluate varKoding, a genome-wide k-mer fingerprinting approach that classifies low-coverage sequencing data without genome assembly using 296 specimens representing seven species of the An. gambiae complex and two non-complex Anopheles species. Leave-one-out validation was performed across five sequencing data amounts (500K–10M bp) using varKode and conventional chaos game representation (CGR) images. varKode consistently outperformed CGR at every sequencing depth, with multi-class accuracy increasing from 95.27% at 500K bp to 98.65% at 10M bp. At 10M bp, six of nine taxa achieved 100% recall; An. gambiae s.s. and An. coluzzii each achieved 97.5% recall, while An. stephensi achieved 95.0%. Reciprocal gambiae–coluzzii errors were concentrated among specimens from Tiassalé, Côte d'Ivoire, including a pre-2013 specimen whose species label cannot be independently verified. Confidence scores generally separated correct from incorrect predictions. At 10M bp, every specimen of An. arabiensis, An. quadriannulatus, An. melas, An. merus and An. bwambae (complex members) and An. funestus (outgroup) was correctly classified, with residual misclassification largely confined to the closely related An. gambiae–An. coluzzii pair.Overall, varKoding provides high-accuracy identification from ultra-low-coverage data across the represented members of the complex, including reliable identification of the important malaria vector An. arabiensis and the non-vector An. quadriannulatus. Difficult specimens can be flagged for orthogonal molecular or population-genomic confirmation, allowing the method to serve as a practical first-pass surveillance tool. Independent geographic validation and broader reference representation would further extend this approach to routine surveillance.