Artificial intelligence can help optimize fruit cultivars to match consumer taste preferences

Which flavors and chemical compounds make a particular variety of fruit more appealing to consumers can be identified and predicted using artificial intelligence.

Breeding for flavor is a difficult task for many different reasons. For one, fruit and vegetable plant breeding programs need to improve several different traits that appeal to both producers and consumers.

Creating the optimal genetic combination that covers all these traits is difficult, so breeding programs often deprioritize flavor to focus on improving disease resistance and increasing yield. Plant breeders must also evaluate hundreds to thousands of potential varieties. Testing a single sample in an objective way requires consumer panels of up to 100 people, which can be expensive and impractical to arrange.

To streamline this process, we developed an algorithm to predict how consumers will rank flavor in tomatoes and blueberries. We created a database containing all known compounds associated with flavor in all varieties of these fruits.

Then, we compared this database with existing consumer panel ratings on sweetness, sourness, umami and overall flavor and preference of different varieties. By modeling how consumer ratings varied with the chemical makeup of different varieties of these fruits, this allowed us to determine which compounds most influence flavor perception.