Determining mosquito age using surface-enhanced Raman spectroscopy and artificial neural networks: insights into the influence of origin and sex

Abstract

Background: Mosquito-borne diseases, such as malaria, dengue, and Zika, continue to pose significant threats to global health, resulting in millions of cases and thousands of deaths each year. Notably, only older mosquitoes can transmit these diseases. Therefore, accurate age estimation of mosquitoes is vital for targeted interventions and risk assessments. However, traditional methods, such as tracheole morphology analysis, are labor-intensive and have limited scalability. Surface-enhanced Raman spectroscopy (SERS), when coupled with artificial neural networks (ANNs), offers a robust and flexible alternative, facilitating accurate and efficient mosquito age determination even in diverse and complex environmental conditions.

Methods: We analyzed 124 Aedes aegypti mosquitoes from California (CA) and Thailand (TH) using SERS, each generating 20 spectra. The ANNs utilized a multilayer perceptron with two hidden layers of 100 neurons and rectified linear unit (ReLU) activation. Classification tasks used cross-entropy loss; regression applied mean squared error. Models were trained with a 70-30 training-validation split and optimized using the Adam optimizer over 10,000 iterations. Performance metrics included accuracy, correlation coefficient (R), and root mean square error (RMSE). t-Distributed stochastic neighbor embedding (t-SNE) visualizations and confusion matrices offered additional model insights into effectiveness.

Results: The ANN models demonstrated superior performance in differentiating mosquito age relative to non-ANN methods. For female CA mosquitoes, the models classified ages from day 1 to day 21 with 84% accuracy and predicted age with an R of 0.96 and RMSE of 2.18 days. Similarly, the models achieved 86% accuracy and an R-value of 0.95 for female TH mosquitoes. While mosquito origin and sex influenced performance, the combined model maintained robust results, achieving 80% accuracy and an R-value of 0.93. Implementing a voting mechanism across multiple spectra for each mosquito significantly improved accuracy, increasing classification performance from approximately 80% at the spectrum level to 100% at the mosquito level.

Conclusions: This study demonstrates the effectiveness of SERS combined with ANN for accurate age classification and prediction of Ae. aegypti mosquitoes. The models achieved high accuracy across diverse populations, with a voting mechanism enhancing classification to 100%. These findings highlight the potential of SERS-ANN as a reliable tool for vector control and disease surveillance.

Keywords: Age prediction; Artificial neural networks; Mosquito classification; Surface-enhanced Raman spectroscopy.

Fig. 1

Sample acquisition map and the design and configuration of the ANN model. A Sample acquiring area (red square) and 10 sub-areas (white and yellow rectangles) used for data collection. B ANN design and configuration for age classification and regression

Fig. 2

Age classification and regression analysis for mosquitoes collected for the CA-origin. A Average SERS spectra of mosquitoes collected on days 1, 7, 14, and 21 at the CA-origin. B Confusion matrix showing the accuracy of the model in classifying mosquito samples by collection day. C t-SNE plot illustrating the clustering of mosquito samples based on collection day. D Regression analysis comparing the actual versus predicted collection days. E Predicted mosquito age (mean ± standard deviation)

Fig. 3

Age classification and regression analysis for mosquitoes collected for the TH-origin. A Average SERS spectra of mosquitoes collected on days 3, 10, and 18 at the TH-origin. B Confusion matrix showing the accuracy of the model in classifying mosquito samples by collection day. C t-SNE plot illustrating the clustering of mosquito samples based on collection day. D Regression analysis comparing the actual versus predicted collection days. E Predicted mosquito age (mean ± standard deviation)

Fig. 4

Days classification and regression analysis for mosquitoes collected from both origins. A Confusion matrix showing the accuracy of the model in classifying mosquito samples by collection day. B t-SNE plot illustrating the clustering of mosquito samples based on collection day. C Regression analysis comparing the actual versus predicted collection days. D Predicted mosquito age (mean ± standard deviation)