A neural network prediction of environmental determinants of Anopheles sinensis knockdown resistance mutation to pyrethroids in China. Issue 2 (17th November 2016)

• Record Type: Journal Article
• Authors: Wei, Xing; Yan, Guiyun; Zhou, Guofa; Zhong, Daibin; Fang, Qiang; Yang, Xiaodi; Hu, Dehua; Chang, Xuelian
• Is Part Of: Journal of vector ecology. Volume 41:Issue 2(2016:Dec.)
• Page Start: 295
• Page End: 302
• Publication Date: 2016-11-17
• Subjects: Anopheles sinensis -- insecticide resistance -- knockdown resistance frequency -- neural network model -- environmental determinants; Arthropod vectors -- Ecology -- Periodicals; Animals as carriers of disease -- Periodicals; Animal ecology -- Periodicals; Host-parasite relationships -- Periodicals; Pests -- Control -- Periodicals; 571.986
• DOI: 10.1111/jvec.12226
• Languages: English
• ISSNs: 1081-1710

Abstract

ABSTRACT: Selection pressure caused by long‐term intensive use of insecticides is the key driving force in resistance development. Additional parameters such as environmental conditions may affect both the mosquito response to insecticides and the selection of resistance mechanisms. In this context, we analyzed the environmental determinants of kdr prevalence in Anopheles sinensis across China. We collected kdr frequency from 48 sites across central and southern China, together with key environmental factors including long‐term climatic data, topographic features, main crops, and land cover types. Trend surface analysis found that the distribution of kdr frequency can be partitioned into three regions, namely central China ( kdr frequency >80%), western China ( kdr frequency varies from 0% to 60%), and southern China ( kdr frequency <10%). Seven predictor variables were selected based on a radial basis function neural network model. A multilayer perceptron (MLP) network model revealed that the number of crops in a year was the most important predictor for the kdr mutation rate. Topography, long‐term mean climate and land cover all contributed to the kdr mutation rate. The observed mean kdr frequency was 53.0% and the MLP network model‐predicted mean was 52.6%, a 0.1% relative error. Predicted kdr frequencies closely matched the observed values. The model explained 92% of the total variance in kdr frequency. The results indicated that kdr was associated with the intensity of pesticide usage. Crop cultivation information, together with environmental factors, may well predict the spatial heterogeneity of kdr mutations in An. sinensis in China.