From minutes to bounds: A probabilistic UV-C control and a shape-only morphological fingerprint for postharvest colletotrichum inactivation in cacao and coffee processing

Authors: Ahn, Ezekiel; Baek, Insuck; LIM, SEUNGHYUN - Orise Fellow; Lovelace, Amelia; CHA, MINHYEOK - Orise Fellow; Kim, Moon; Park, Sunchung; Meinhardt, Lyndel

Submitted to: Food Control
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/2/2026
Publication Date: 1/2/2026
Citation: Ahn, E.J., Baek, I., Lim, S., Lovelace, A.H., Cha, M., Kim, M.S., Park, S., Meinhardt, L.W. 2026. From minutes to bounds: A probabilistic UV-C control and a shape-only morphological fingerprint for postharvest colletotrichum. Food Control. 183. Article 111956. https://doi.org/10.1016/j.foodcont.2026.111956.
DOI: https://doi.org/10.1016/j.foodcont.2026.111956

Interpretive Summary: UV-C light (ultraviolet light) is a popular, chemical-free method for killing mold on fruits like cacao and coffee, but the common "10-minute" rule isn't always reliable. Different strains of fungi exhibit vastly different levels of resistance, and our study found that one remarkably resilient strain (P24-88) could still achieve a 100% survival rate even after a 10-minute treatment. To address this safety issue, we have developed a new, two-part system. First, we use physics and statistics to replace the guesswork of "minutes" with an actual "probabilistic guarantee" of how many fungi are safely killed. Second, we discovered that a fungus's "shape"—not its size—changes in a specific way when it's stressed. We utilized artificial intelligence (AI) to learn this "shape fingerprint," and it was able to identify the origin of a fungus (a cacao or coffee farm) with 93% accuracy. This research provides a framework for building smarter, more reliable UV sterilization systems based on proven risk levels, rather than just time, which helps make our food supply safer. This research empowers food safety engineers, equipment manufacturers, and regulatory agencies to abandon unreliable "time-rules" and instead design smarter sterilization systems based on precise risk calculations, ultimately ensuring a safer food supply for consumers.

Technical Abstract: Standard time-based physical decontamination protocols (e.g., UV-C for ~10 min) often fail to provide statistical safety guarantees due to significant biological heterogeneity among pathogens. To address this critical gap in postharvest process control for high-value crops like cacao and coffee, we present a dual framework, validated under controlled in vitro conditions, that substitutes empirical heuristics with probabilistic lethal bounds and a rapid morphological diagnostic. We utilized a large Colletotrichum dataset (n = 5363) spanning diverse treatments (UV-C, UV-B, sonication) to validate this approach. First, analysis of standard UV-C treatments (275 nm, ~348 mJ cm-2) revealed that average-based protocols mask “tail-risk” hazards; the most conservative Clopper–Pearson upper 95 % bound on survival reached 1.000 for tolerant isolates (e.g., P24-88), indicating that survival probability could reach 100 % at the 95 % confidence level and highlighting a failure of the fixed-time process to ensure universal control. Second, we developed a 'shape-only' machine learning framework (Gradient Boosting) to facilitate rapid process verification. Using plate-grouped cross-validation (Group K Fold) to ensure robust assessment, this morphological classifier successfully predicted pathogen host-origin (Accuracy ˜ 0.93) and the continuous survival ratio (treated/control area) (R2 ˜ 0.74), serving as a non-destructive indicator of UV-induced physiological stress. This work bridges the gap between fungal biology and food engineering, offering a robust Quality Assurance (QA) tool that transforms physical decontamination from an empirical art into a validated, statistically controlled engineering process.