Overall pipeline of our approach. The three steps respectively involve obtaining pre-trained embeddings for single molecules as upstream information, using exponential transformation to derive the embedding for mixtures, and finally employing a boosting method to learn the perceptual distance between olfactory mixtures.

Congratulations to the student team of Yikun Han (MDS), Zehua Wang (MDS alum), and Stephen Yang (undergrad), and to their faculty advisor Prof. Ambuj Tewari who placed top (in a remarkable 4-way tie for 1st place) in the international DREAM challenge on olfactory mixture prediction, where their task was to predict how similar two mixtures of odorants will smell.   

Representing the winning team, Yikun will present this work at the RSGDREAM 2024 conference. Congrats all!