Abstract
AbstractAccurate identification of disease vector insects is crucial when collecting epidemiological data. Traditionally, mosquitoes that transmit diseases like malaria, yellow fever, chikungunya, and dengue fever have been identified by looking at their external morphological features at different life cycle stages. This process is tedious and labour intensive.In this paper, the potential of Raman spectroscopy in combination with Linear and Quadratic Discriminant Analysis to classify three mosquito species, namely: Aedes aegypti, Anopheles gambiae and Culex quinquefasciatus, was explored. The classification was based on the mosquitoes’ cuticular melanin. The three mosquito species represented two subfamilies of medically important mosquitoes, i.e. the Anophelinae and the Culicinae. The housefly (Musca domestica) was included as a ‘control’ group to assess the discrimination ability of the classifiers. This study is the first to use Raman spectroscopy to classify mosquitoes. Fresh mosquitoes were anaesthetized with chloroform, and a dispersive Raman microscope was used to capture spectra from their legs. Broad melanin peaks centred around 1400 cm-1, 1590 cm-1, and 2060 cm-1 dominated the spectra. Variance Threshold (VT) and Principal Component Analysis (PCA) were used for feature selection and feature extraction respectively from the preprocessed data. The extracted features were then used to train and test Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA) classifiers.The VT/PCA/QDA classification model performed better than VT/PCA/LDA. VT/PCA/QDA achieved an overall accuracy of 94%, sensitivity of 87% and specificity of 96%, whereas VT/PCA/LDA attained an accuracy of 85%, a sensitivity of 69% and a specificity of 90%. The success of these relatively simple classification models on Raman spectroscopy data lays the groundwork for future development of models for discriminating morphologically indistinguishable insect species.
Publisher
Cold Spring Harbor Laboratory