Investigation of aromatic compounds and olfactory profiles in cocoa pulp fermentation using yeast-based starters: A Volatilomics and machine learning approach

Abstract

The interaction and complex metabolism of microorganisms in cocoa pulp drive the fermentation process. To investigate this, four strains from spontaneous cocoa fermentation, including Hanseniaspora uvarum, Saccharomyces cerevisiae, Lactiplantibacillus plantarum, and Gluconobacter potus were combined to ferment cocoa pulp. Nineteen machine learning algorithms were run with the dataset of volatile compounds quantified by headspace solid-phase microextraction gas chromatography–mass spectrometry (HS-SPME-GC–MS) against integrated olfactory evaluation to reveal metabolite-sensory attribute relationships. The models showed high prediction accuracy, ranging from 0.85 for sourness by Gradient Boost Machine to 0.28 for sweetness by linear regression. Ethyl esters, specifically ethyl octanoate and ethyl 9-decenoate, were found positive for aroma development. Polynomial regression, neural network modeling and gradient boosting decision trees highlighted the high carbohydrate consumption rate of S. cerevisiae, the pectin degradation ability of H. uvarum, and the synergy of lactic acid bacteria with G. potus. This study offers new insights into cocoa flavor and the development of fermentation starter cocktails.