Work

Machine learning-based volatile profiling to explore the potential characteristics of pest resistance in different Castanea mollissima cultivars

Year: 2025

Type: article

Abstract: As an important woody food crop, Castanea mollissima are rich in nutrients and functional components. However, the yield of C. mollissima fruits is seriously affected by pest infestations in several production areas of China. Herein, six C. mollissima cultivars with different pest rates were selected to explore the potential reasons by combining with volatile profiles and machine learning analysis. The results indicated that E-nose demonstrated the effective discriminative ability, with the W1W, W2W, W1S and W5S sensors exhibiting the strongest responses across all C. mollissima bur samples. Among the all identified volatile organic compounds (VOCs), alcohols and aldehydes were the main VOCs in every sample across the three developmental stages (July 1st, July 25th, August 12th). Thereinto, the volatile characteristics of C. mollissima burs at first two stages were highly correlated with the actual pest infestations. Based on the GC–MS datasets from the first two stages, six machine learning models were developed to evaluate their performance in capturing and learning the volatile characteristics of the samples. Specially, the ANN model displayed the best learning and generalization capabilities than the other models. Ten key VOCs (i.e., (E)-4,8-dimethylnona-1,3,7-triene, hexanal, (E)-2-hexenal, Z-3-hexenol, 1-hexanol, eugenol, 1-nonanol, 1,8-cineole, geraniol, 1-octanol) were identified by ANN model, which exhibited the strong potential in distinguishing the different C. mollissima samples. The possible reasons of the differential insect resistance in these C. mollissima cultivars were clarified successfully. These information would provide a theoretical foundation for developing the pest-resistant C. mollissima cultivars and establishing the sustainable pest management strategies.

Source: Scientia Horticulturae

DOI: 10.1016/j.scienta.2025.114552