Improved Ensemble\nModel for Insecticide Recognition\nby Incorporating Insect Toxicity Data

Pesticide molecules, such as insecticides, play a critical\nrole\nin modern agricultural production. Traditional pesticide development\nmethods are often inefficient and expensive, while data-driven artificial\nintelligence (AI) techniques have emerged as a useful tool to facilitate\ndrug discovery. However, currently available commercial pesticide\ndata is limited, which makes the trained models unsatisfactory in\nterms of performance and generalization. From a domain knowledge perspective,\ninsect toxicity data were incorporated to improve the insecticide\nrecognition of AI models. Compared to the models trained with the\noriginal data set, the new models performed better in the external\nvalidation, and their generalization was more desirable. In addition,\nby integrating different types of individual models, we obtained an\nensemble model with better performance. Based on this, an online platform\nwas developed to provide researchers with free access to insecticide\nscreening (https://dpai.ccnu.edu.cn/InsectiVS/). Finally, two potential insecticide molecules with insecticidal\nactivity against <i>Plutella xylostella</i> were successfully\nidentified in a real-world scenario. In conclusion, this idea connects\nthe fields of AI and agricultural chemistry and is expected to have\nwide application in pesticide research.