QSAR modeling and machine learning for antifungal peptide discovery in crop protection: Translational challenges and future perspectives

Abstract

Fungicide resistance and restrictions on conventional agrochemicals are increasing the need for sustainable crop-protection strategies. Antifungal peptides (AFPs) are considered promising candidates due to their multiple mechanisms of action, biodegradability, and suitability for environmentally sustainable disease control. However, their application in agriculture remains limited by poor field stability, suboptimal delivery systems, high production costs, and the lack of predictive models validated against phytopathogenic fungi. This review collected and critically compared QSAR and machine learning (ML) studies published between 2015 and 2025, focusing on their modeling strategies, predictive performance, limitations, and translational requirements for antifungal peptide discovery in crop protection. The available literature remains dominated by Candida spp. and broadly biomedical antifungal datasets, highlighting the translational gap between biomedical AFP prediction and agricultural application. Eight representative studies were compared in terms of dataset size, endpoint, descriptors, algorithms, target fungi, predictive performance, and validation approaches. Dataset size ranged from 45 to 60 peptides in small congeneric series to large-scale collections containing up to 17,514 sequences. Classification models generally reported validation or test accuracies of 0.86–0.96 and AUC values of 0.91–0.99, whereas regression models showed more variable external or test performance, with R² values of approximately 0.50–0.87. The reviewed studies illustrate a diverse methodological landscape, from interpretable QSAR models to ML classifiers and deep-learning pipelines based on protein-language-model sequence representation. Overall, the principal barrier to agricultural translation is not algorithm availability, but the lack of curated plant-pathogen-specific datasets, standardized agricultural endpoints, and external benchmarks incorporating field-relevant traits such as protease resistance, UV stability, leaf-surface persistence, formulation compatibility, phytotoxicity, and effects on beneficial microorganisms.