PMC

2D t-SNE projection of the 300-D embeddings of 40 food ingredients, five from each of the eight food categories. (A) FlavorDB (B) FlavorGraph.

The figure shows flavor preference response patterns in the 5-class latent class analysis solution. All participants were asked, “How much do you like the following flavors for e-cigarettes or vapes?” for eight listed flavors (Fruit flavors; Candy or dessert flavors; Mint flavors; Icy, frost, or menthol flavors; Combined icy flavors, like “icy fruit”; Alcohol flavors; Tobacco flavor; Unflavored) with response options: strongly dislike; somewhat dislike; neither like or dislike; somewhat like; strongly like. The height of the unfilled bars shows the percent of the total sample that endorsed each option for each flavor. The height of the filled bars shows the percent of the members assigned to the specific class that endorsed each option for each flavor. For example, members of Class 1 were less likely than the total sample to endorse strongly dislike, somewhat dislike, and neither like or dislike for the flavor mint, while also being more likely than the total sample to endorse strongly like for mint.

FlavorDB is a seamless amalgamation of ‘entity space’ and ‘flavor space’. The resource provides a comprehensive dataset along with a user-friendly interface and interlinked search engines for exploring the flavor universe.

Effects of E-liquid ± high and low percent flavorings in BHI on planktonic bacterial growth. Each bar represents Mean ± SEM percent change from the planktonic growth in BHI control media (no E-liquids), n = 4 to 12 are the number of replicates. a = p < 0.05 from flavorless E-liquid, b = p < 0.01 from flavorless E-liquid and c = p < 0.001 from flavorless E-liquid.