Abstract
According to the Food and Agriculture Organization of the United Nations (FAO), pests reduce global crop production by 14% annually. The growing challenge of pest resistance, coupled with the relatively low success rates of pesticides, has prompted researchers to shift their attention towards the accurate evaluation of insecticide lead. In contrast to in vitro methods of structural similarity or target affinity, the 'insecticide-likeness' approach emphasises the in vivo biological effects of compounds, thereby constructing precise and comprehensive evaluation rules. In the present study, a multi-scale qualitative-quantitative insecticide-likeness evaluation platform, Agrochem Predictive Platform for Insecticide-likeness (APPi), was developed. An APPi rule was proposed for qualitative evaluation (ClogP ≤ 7, ARB ≤ 18, HBA ≤ 7, HBD ≤ 2, PFI ≤ 8 and ROB ≤ 10). A quantitative insecticide-likeness evaluation model, the APPi model, was developed based on a multi-classifier integrated machine learning framework (PUMV). The APPi model demonstrated excellent performance on the train and external test sets. Crucially, on the independent external test set, it achieved an accuracy of 85%, which represents a significant improvement over existing models. Furthermore, we developed the FragScore Visualiser tool to identify critical insecticidal fragments of compounds. The APPi platform provides precise guidance for virtual screening and structure optimisation of lead compounds in the early stage of insecticides discovery. The platform is available free of charge at http://pesticides.cau.edu.cn/APPi.