Development of an AI Method Based on Chemical Fingerprints to Discriminate Defective Cocoa Liquors

Theobroma cacao represents one of the most economically important commodities in the world and is valued worldwide for its sensory properties and potential health benefits, including cardiovascular effects and mood enhancement [1]. However, climate change significantly impacts cocoa quality, and political instability in producing regions and around the world have exacerbates the cocoa market situation [2]. So, ensuring cocoa quality remains a challenge due to variability in chemical composition and sensory properties influenced by environmental and processing factors. This study, in collaboration and supported by Ferrero Italia, aims to develop an AI-based method using chemical fingerprints to objectively discriminate between defective and high-quality cocoa liquors [3]. Cocoa quality control involves assessing the intermediate products (liquor) flavor through sensory tests to determine their quality. During this phase, several samples were found to no longer comply with industry reference standards due to the presence of off-flavors. Omics approach has been employed to characterize the issue and define acceptance specifications for input products. [4-5] The resulting chromatographic fingerprint were processed combining machine learning tools and Data Fusion methods to identify chemical-sensory differences between defective samples and quality standards [3-6]. This integrated analytical approach, combining multiple chemical fingerprinting techniques with machine learning algorithms, offers a powerful tool for enhancing quality control in the cocoa supply chain. By leveraging advanced analytical and computational capabilities, it goes beyond traditional industry-defined marker specificity, enabling a reliable identification of off-flavor samples even in cases of high variability in sensory descriptions. This methodology not only ensures consistent product quality despite environmental challenges but also strengthens the detection of subtle yet critical deviations that may impact the final product.