Understanding the effects of sub-lethal exposure to bioactive compounds is crucial for advancing both toxicological assessment and biosensing systems. In this study, an artificial intelligence (AI)-driven behavioral phenotyping framework was used to investigate sex-specific responses in the Mediterranean fruit fly Ceratitis capitata after ingestion of an LC30 concentration of carlina oxide. The considered features were derived from deep learning-based markerless pose estimation of video recordings and characterized locomotion, wing activity, and rotational dynamics. Statistical analysis revealed distinct sexrelated behavioral differences, which were more widespread under treatment conditions. Machine learning classification further supported these findings, with Random Forest (RF) achieving the highest performance in the treated male–female comparison (accuracy = 0.79, F1-score = 0.80, ROC-AUC = 0.87). Treatmentrelated behavioral differences were more evident in males than in females, with a ROC-AUC of 0.71 for the comparison between control and treated males. These results demonstrate that sublethal exposure is associated with measurable sex-dependent behavioral differences that can be effectively captured through AI-based analysis. The proposed framework provides a scalable and non-invasive tool for behavioral toxicology, with potential applications in toxicological assessment, pest management, and the development of bio-hybrid and biosensing platforms.
Using Machine Learning Techniques for Behavioral Phenotyping in a Fruit Fly Reveals Sexual Dimorphism Following Sub-Lethal Insecticide Exposure
Manduca, Gianluca;Casadei, Anita;Romano, Donato
2026-01-01
Abstract
Understanding the effects of sub-lethal exposure to bioactive compounds is crucial for advancing both toxicological assessment and biosensing systems. In this study, an artificial intelligence (AI)-driven behavioral phenotyping framework was used to investigate sex-specific responses in the Mediterranean fruit fly Ceratitis capitata after ingestion of an LC30 concentration of carlina oxide. The considered features were derived from deep learning-based markerless pose estimation of video recordings and characterized locomotion, wing activity, and rotational dynamics. Statistical analysis revealed distinct sexrelated behavioral differences, which were more widespread under treatment conditions. Machine learning classification further supported these findings, with Random Forest (RF) achieving the highest performance in the treated male–female comparison (accuracy = 0.79, F1-score = 0.80, ROC-AUC = 0.87). Treatmentrelated behavioral differences were more evident in males than in females, with a ROC-AUC of 0.71 for the comparison between control and treated males. These results demonstrate that sublethal exposure is associated with measurable sex-dependent behavioral differences that can be effectively captured through AI-based analysis. The proposed framework provides a scalable and non-invasive tool for behavioral toxicology, with potential applications in toxicological assessment, pest management, and the development of bio-hybrid and biosensing platforms.| File | Dimensione | Formato | |
|---|---|---|---|
|
Using_Machine_Learning_Techniques_for_Behavioral_Phenotyping_in_a_Fruit_Fly_Reveals_Sexual_Dimorphism_Following_Sub-Lethal_Insecticide_Exposure.pdf
accesso aperto
Tipologia:
Documento in Pre-print/Submitted manuscript
Licenza:
Creative commons (selezionare)
Dimensione
1.49 MB
Formato
Adobe PDF
|
1.49 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

