Machine learning classification of semiochemicals using structural and complexity features

Abstract The relationship between semiochemical activity and molecular complexity (MC) was investigated using machine learning (ML). Using literature reported data on the electroantennogram response of the red-necked long horn beetle to semiochemicals, ML classification models were created using MC and other related descriptors as the model features. The created ML models revealed a non-linear relationship between MC and semiochemical activity, suggesting an optimum threshold for MC. This implies that immensely complex molecules do not translate into better semiochemical activity. The contribution of MC – based descriptors was likewise quantified by calculating their respective Shapley values, wherein the sex-based differences in insect olfaction were satisfactorily captured by the models. Taken together, the study presents preliminary but encouraging results that can guide future work and experiments that can further enhance our understanding of the nuances of insect olfaction and semiochemistry.