Towards Scalable Age-Grading of Aedes albopictus mosquito using Mid-Infrared Spectroscopy and Machine Learning

Abstract

The age structure and dynamics of mosquito populations are crucial for understanding their ability to spread diseases and assessing the effectiveness of anti-mosquito control measures. However, available methods to age-grade mosquito populations are labour-intensive and imprecise, particularly for Aedes species. We investigated the potential of Mid-Infrared Spectroscopy (MIRS) combined with Supervised Machine Learning (ML) to rapidly and accurately predict the age of adult females and males of the arbovirus vector, Aedes albopictus. First, we demonstrated the ability of MIRS-ML to age male and female mosquitoes reared under laboratory conditions. Second, we optimised the model with adults emerged from wild collected eggs reared under natural conditions in a semi-field facility, to expose them to more realistic ambient conditions. For each sex we developed three ML models based on the resolution of the predicted adult age class: low (9 day interval), medium (6 days) and high resolution (3 days) from 1 to 15 or 33 days for males and females, respectively. The prediction accuracy decreased as the resolution increased. In males, the accuracy dropped from 99% (low) to 93% (medium) and 85.8% (low); in females the high and medium resolution models showed 89.4% and 78.5% accuracy, which decreased to 72.6% for the low resolution. In a simulated vector control intervention, the low-resolution models allowed to detect shifts in the age-structure of Ae. albopictus populations with minimal sampling effort (<100 specimens). Finally, we validated MIRS-ML on two unseen data and reconstructed plausible age structures in 1) laboratory-reared and 2) field-collected Ae. albopictus males and females. Overall, the results represent a first step towards the development of a sound and reproducible MIRS-ML approach for age-grading of Ae. albopictus populations in the wild. AUTHOR SUMMARYKnowing the age of mosquito populations is critical for understanding how effectively they can transmit viruses like dengue, chikungunya, and Zika, as older mosquitoes are more likely to be infectious. Also, comparing ages of mosquito population before and after a control intervention - such as insecticide aerial spraying - may allow to understand the impact of the intervention. However, existing methods to estimate mosquito age are time-consuming and imprecise. In this study, we tested whether a rapid and scalable method based on detection of age-related changes by mid-infrared spectroscopy (MIRS) coupled with machine learning (ML) could accurately estimate the age of Aedes albopictus, the Asian Tiger mosquito, an important arbovirus vector. We trained MIRS-ML models using mosquitoes reared in both laboratory and semi-field conditions to reflect realistic environmental variation. Our models were able to classify mosquito age with high accuracy, especially when grouping individuals into broader age categories. In simulated vector control scenarios, low-resolution models effectively detected shifts in population age structure with minimal sampling effort. We also applied our approach to field-collected mosquitoes that showed plausible age structures, suggesting potential of this approach for real-world surveillance. This method represents a promising, scalable, and non-destructive tool for monitoring mosquito population dynamics and could help monitor control strategies against Aedes-borne diseases.