Tomatoes, a widely cultivated vegetable, are susceptible to infections by Alternaria species, which can lead to contamination with mycotoxins such as alternariol (AOH), alternariol monomethyl ether (AME), and tenuazonic acid (TeA), posing public health risks. This study presents Altosafe , a predictive model combined with a Random Forest approach for exploratory assessment of Alternaria mycotoxin contamination in open-field tomatoes. First, a cumulative fungal infection index was constructed based on literature and experimental data. This cumulative fungal infection index was then integrated with weather and agronomic data collected from open-field tomato cultivation trials held in China to develop predictive models using the random forest (RF) algorithm. Field trial data collected between April and November 2023 were used for model development and internal validation, while samples collected in May and June 2023 were reserved for hold-out validation within the same experimental context. Separate RF classification models were developed for AOH, AME, and TeA using predefined contamination thresholds. Internal validation accuracies ranged from 96.0% to 98.0%, and hold-out validation accuracies ranged from 70.0% to 82.0%. Threshold adjustment demonstrated expected trade-offs between sensitivity and specificity. To our knowledge, this is the first study in China to explore predictive models for Alternaria mycotoxin contamination in tomatoes. While the models showed promising performance under constrained field conditions, broader multi-season and multi-location validation is required before wider application can be considered. • Altosafe integrates a cumulative fungal infection index with random forest approach. • Altosafe models are based on simplified biological assumptions and limited data. • Fungal infection risk index is the most important predictor among all inputs. • Hold-out validation accuracies between 74% and 82% in one field context. • AltoSafe models provide Alternaria mycotoxin contamination risk at set thresholds.
Zhang et al. (Wed,) studied this question.