Effectively saltiness enhanced odorants screening and prediction by database establish, sensory evaluation and deep learning method

Odor-taste interaction has gained success in enhancing saltiness perception. This work aimed to provide candidate odorants for saltiness enhancement. Volatile compounds and their frequencies in salty foods were systematically analyzed. The compounds with higher frequency were incorporated into the savory aroma compounds database. The saltiness enhancement concentrations of representative aroma compounds at the NaCl solution (3.00 g/L) were detected by sensory evaluation. SELF-referencing Embedded Strings-based representation leaning and graph attention network combined with Backpropagation Neural Network classifier was utilized to predict the saltiness-enhancing ability of odorants. Results showed that ketones, pyrazine and sulfur-containing compounds showed higher saltiness-enhancing ability. Mushroom and fatty attributes contributed to the saltiness-enhancing ability of aroma compounds. Deep learning model showed excellent generalization ability and accuracy (95.93 %), which provided rapid screening method for selecting savory aroma compounds. This study would provide new pathways for food industry to achieve salt reduction goals.