Recent advances in predictive methods and availability of perceptual data have paved the way for a growing interest in olfactory perception predictions from chemical representations of molecules. This has led to a growing consensus that for pure odors, it is possible to build models using the chemical structure of molecules to predict the perceptual values of natural language attributes of smells. However, predictions have mainly focused on pure molecules and not the real-world situation of olfactory mixtures. In order to start filling this gap, we plan to organize a second DREAM olfaction prediction challenge now focused on predicting the discriminability of olfactory mixtures. Using publicly available data from 3 different studies (Bushdid et al 2014, Snitz et al 2013, Ravia et al 2020) for more than 700 unique mixtures and almost 600 measurements of mixture pairs discriminability, participants will be tasked to predict the discriminability of 46 unpublished mixture pairs. We will here present the details of the datasets and the challenge timeline/scoring approach.
