Transfer Learning Approach to Develop Natural Molecules with Specific Flavor Requirements

In the past decades, the flavor industry's investmentinresearch and development has increased to take innovative steps. Meanwhile,the lack of information regarding the flavored molecules and specificflavoring properties is an obstacle to advances in this sector. Inthis context, this work presents the implementation of three scientificmachine learning techniques as an innovative methodology to designnew natural flavor molecules with specific desired properties to productdevelopment. The transfer learning technique is presented to tacklethe lack of data available when analyzing flavor molecules. Nine flavordescriptors were studied along this work, and all of them presentedmore than 50% of molecules generated within the outstanding resultsconsidered for the evaluation metric, natural product-likeness scoreand synthetic accessibility score. Finally, a discussion of the resultsis constructed based on the data availability, the presence in nature,and the multisensorial flavor component impact for the specific flavors'results.