Predicting the antifungal effects of ethanolic propolis and geopropolis extracts on Aspergillus flavus growth using...

This study evaluated the antifungal potential of ethanolic propolis (EEP) and geopropolis (EEG) extracts against Aspergillus flavus, integrating mechanistic modelling and machine learning (ML) approaches to predict growth kinetics. Chemical characterization showed that EEG contained a gallic acid concentration ca. 80-fold higher than EEP (12.88 vs. 0.16 mg/g), whereas EEP exhibited higher total phenolic (129.88 mg GAE/g) and flavonoid (34.31 mg QE/g) contents. Both extracts displayed dose-dependent fungistatic effects. At 10% (v/v) EEP, the maximum growth rate (μ_max) of A. flavus, predicted using a one-step model (Gibson primary model coupled with an exponential-type secondary model), decreased from 13.69 to 8.87 mm/d. The inhibitory coefficient (k) estimated by the model was slightly higher for EEG (0.05% v/v) than for EEP (0.04% v/v), indicating a slightly stronger antifungal effect of EEG. Fungal growth kinetics were adequately described using the mechanistic modelling approach. However, ML approaches, including Support Vector Regression (SVR), Random Forest Regression (RFR), and Gaussian Process Regression (GPR), demonstrated superior predictive performance. Among them, GPR achieved the highest accuracy for both extracts, with RMSE values <5 mm and adjusted R² > 0.98, outperforming traditional kinetic models. The results demonstrate that EEP and EEG represent promising natural alternatives to synthetic fungicides for controlling A. flavus in food, particularly during and post-harvest systems of stored commodities such as grains, cereals, and nuts. Furthermore, translating these predictive models into accessible software tools offers robust decision-support systems to optimize multi-factor preservation strategies and guide the effective application of natural antifungal extracts in food systems, contributing to more sustainable and effective preservation strategies. However, their application in real food matrices remains to be validated for food industry applications.