Each year, fragrance companies develop and patent two or three new molecules called captives. 1 These molecules are created not only to entice consumers with unique notes for perfumery but also to decrease cost, allergic reactions, feedstock shortages, or non-biodegradability/non-renewability of previous raw materials.To launch two to three captives per year on the market, screening of hundreds or thousands of molecules is necessary. As such, fragrance design requires considerable experimental work of synthesis and characterization. Hence, predicting in silico the physicochemical properties of virtual candidates can significantly contribute to a faster development of new captives with targeted features.According to the REACH regulation, 2 17 properties must be determined to bring a new molecule to the market (Table S1). However, measuring experimentally all these properties is not always realistic in terms of time, cost, feasibility at R&D level, and safety (notably for hazardous compounds). Therefore, this implementation has motivated the development of alternative approaches to experimental testing with fast and reliable methods. 2 As a consequence, thanks to improvements in computer hardware and software in the past decades, numerous tools to predict property values for large molecule sets have emerged.Perfumery is typically a field of application in which prediction of physicochemical properties can be very useful. Indeed, vapor pressure (VP) and boiling point (BP) as well as Henry's law constant (HLC), which represents the mass transfer of molecules from liquid to air, are essential data to describe and assess the volatility of scent molecules. [3][4][5][6][7] Likewise, water solubility (WS) and octanol-water partition coefficient (log P) can bring valuable insights for the elaboration of aqueous formulations. 4 Among the available predictive approaches, mainly five stand out for estimating chemical properties with their pros and cons. [5][6][7][8][9][10][11] The group contribution model (GC) or additive group method is an empirical method based on the division of molecule in blocks (functional groups). [12][13][14][15] In most cases, the property is computed as a sum