Title: Mos-Gen: A Generative Molecular Framework for Mosquito Insecticide Design

Abstract: Each year, infectious diseases transmitted by mosquitoes result in over 700,000 fatalities globally. The prolonged application of traditional chemical insecticides has led to significant resistance issues, highlighting an immediate necessity for innovative, potent, and environmentally friendly alternatives. Although current artificial intelligence methods in this field largely concentrate on predicting and classifying activity, they fail to address the crucial gap in the de novo creation of new molecular scaffolds. To bridge this divide, we introduce Mos-Gen, a motif-aware generative collaborative system. This framework integrates the pretrained molecular representation model Uni-Mol with a variational autoencoder (VAE), specifically optimized for designing allicin derivatives that contain disulfide bonds for use as mosquito insecticides. From the pool of generated candidates, fourteen compounds were chosen for synthesis and experimental testing; this group included nine predicted positives and five predicted negatives. The experimental outcomes confirmed the framework's high-precision screening ability, with the hit rate among predicted positives standing at 78%, while none of the predicted negatives demonstrated any mosquitocidal activity.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC