Deep learning and targeted metabolomics-based monitoring of chewing insects in tea plants and screening defense compounds

Abstract

Tea is an important cash crop that is often consumed by chewing pests, resulting in reduced yields and economic losses. It is important to establish a method to quickly identify the degree of damage to tea plants caused by leaf-eating insects and screen green control compounds. This study was performed through the combination of deep learning and targeted metabolomics, in vitro feeding experiment, enzymic analysis and transient genetic transformation. A small target damage detection model based on YOLOv5 with Transformer Prediction Head (TPH-YOLOv5) algorithm for the tea canopy level was established. Orthogonal partial least squares (OPLS) was used to analyze the correlation between the degree of damage and the phenolic metabolites. A potential defensive compound, (-)-epicatechin-3-O-caffeoate (EC-CA), was screened. In vitro feeding experiments showed that compared with EC and epicatechin gallate, Ectropis grisescens exhibited more significant antifeeding against EC-CA. In vitro enzymatic experiments showed that the hydroxycinnamoyl transferase (CsHCTs) recombinant protein has substrate promiscuity and can catalyze the synthesis of EC-CA. Transient overexpression of CsHCTs in tea leaves effectively reduced the degree of damage to tea leaves. This study provides important reference values and application prospects for the effective monitoring of pests in tea gardens and screening of green chemical control substances.

Keywords: EC-CA; HCT; prediction model.