Reverse-engineering human olfactory perception from chemical features of odor molecules

摘要: Despite 25 years of progress in understanding the molecular mechanisms olfaction, it is still not possible to predict whether a given molecule will have perceived odor, or what olfactory percept produce. To address this stimulus-percept problem for we organized crowd-sourced DREAM Olfaction Prediction Challenge. Working from large psychophysical dataset, teams developed machine learning algorithms sensory attributes molecules based on their chemoinformatic features. The resulting models predicted odor intensity and pleasantness with high accuracy, also successfully eight semantic descriptors (garlic, fish, sweet, fruit, burnt, spices, flower, sour). Regularized linear performed nearly as well random-forest-based approaches, predictive accuracy that closely approaches key theoretical limit. presented here make perceptual qualities virtually any an impressive degree reverse-engineer smell molecule.