The application of artificial neural networks to the identification of newspinosoids with improved biological activity toward larvae of Heliothis virescens

The spinosyns are a new class of fermentation-derived insect control agents that are effective against a variety of lepidopteran insect pests, includi ng the tobacco budworm Heliothis virescens. Efforts to improve the efficacy of the spinosyns led to synthesis of a variety of spinosoids, synthetic or semisynthetic analogs of the naturally occurring spinosyns. One approach u sed to help define synthetic directions for the spinosoids includes the app lication of artificial neural networks. Artificial neural network-based ana lysis of the naturally occurring spinosyns suggested a number of synthetic directions that could potentially result in spinosoids with improved biolog ical activity compared to spinosyn A. One of the artificial neural network- derived synthetic improvements involved increasing the alkyl chain length o f methoxy groups of the 2',3',4'-tri-O-methylrhamnosyl moiety. The 2',3',4' -tri-O-ethyl analog was predicted to be more active than spinosyn A, and up on synthesis and testing in a variety of bioassays, this was indeed found t o be the case. Depending on the bioassay, the 2',3',4'-tri-O-ethyl analog w as between 3.8 and 13-fold more active than spinosyn A. Further artificial neural network-based analysis suggested that the 3' position would be respo nsible for most of the observed improvement in activity. Again, as predicte d by the neural network models, the 3'-O-ethyl analog was about as active a s the 2',3',4'-tri-O-ethyl analog, while the 2'- or 4'-O-ethyl analogs were much less active. Thus, the use of a novel approach to quantitative struct ure-activity relationships has allowed the identification of new spinosoids with biological activity against larvae of H. virescens superior to that o bserved in the naturally occurring spinosyns. (C) 2000 Academic Press.