Age-grading and species identification of male mosquito Anopheles gambiae s.l. using mid-infrared spectroscopy and machine learning

Most malaria vector control strategies target female Anopheles mosquitoes that transmit parasites, but emerging approaches such as gene drive, sterile insect technique, or Wolbachia require the release of males. Successful deployment depends on reliable knowledge of the species composition and age structure of male populations to overcome potential barriers to mating. Although approaches including near-infrared spectroscopy have been explored for male species identification and age grading, these methods remain constrained by limited validation, scalability, and accuracy under operational field conditions. Here we tested mid-infrared spectroscopy (MIRS) coupled with machine learning (ML) as a rapid and cost-effective approach for classifying species and age of male mosquitoes within the Anopheles gambiae complex. We used male mosquitoes from two laboratory-reared colonies in Burkina Faso (Anopheles coluzzii and An. gambiae) to develop a MIRS-ML model for species and age group (1–4, 5–10, 11–17 days) prediction under controlled conditions. This model was tested against a genetic and environmentally variable dataset consisting of male offspring obtained from gravid or blood fed Anopheles coluzzii and An. gambiae females collected from houses from two villages, Vallée du Kou and Soumousso and reared to adulthood in a semi-field system. The MIRS spectra from 2,120 males, representing both species, all age groups and both laboratory and semi-field backgrounds, were analysed using an extreme gradient boosting (XGBoost) algorithm to assess the ability to correctly predict age group and species. The XGBoost model classified mosquito species (86%) and age groups (85%) accurately in laboratory data, but performance declined under semi-field conditions (64% for species, 50% for age), reflecting environmental variability. Incorporating semi-field samples through transfer learning improved accuracy to 73% for species and 70% for age, underscoring both the limits of laboratory-only models and the value of transfer learning for enhancing generalisability in field settings. Our results demonstrate that mid-infrared spectroscopy with supervised machine learning (MIRS-ML) holds potential as a rapid tool for identifying the species and age group of cryptic male malaria vectors and represent one of the first applications of this approach to male Anopheles gambiae s.l. evaluated under semi-field conditions. However, before the approach can be used, larger datasets are needed to improve the classification algorithms and validate them for prediction in field populations.