A GNN odor-map's geometry matches human perceptual odor similarity

measured in 1 paper

Lee et al. train a message-passing GNN on ~5,000 molecules labeled with 138 human odor descriptors (cross-validated AUROC 0.89), taking the 256-dim penultimate layer as a Principal Odor Map [lee-etal-2023-principal-odor-map] POM pairwise distances correlate R=0.73 with an independently-measured human perceptual odor map, versus R=-0.12 for a Morgan-fingerprint embedding [lee-etal-2023-principal-odor-map] Same-label molecules cluster tighter in the POM (cluster density 0.51 vs 0.68 for fingerprints) [lee-etal-2023-principal-odor-map] On 320 novel odorants rated by a trained panel, the POM model beats the median panelist for 53% of molecules and on 58% of odor descriptors [lee-etal-2023-principal-odor-map] The evidence is an RSA-style distance correlation to perceptual ground truth; no representation-space intervention is performed [lee-etal-2023-principal-odor-map]

Structure

Platonic Representation Hypothesis, Linear Subspace

Context

Message Passing Neural Network (GNN) on molecular graphs, Principal Odor Map (256-dim penultimate-layer embedding), RSA-style distance correlation to human perceptual ground truth (R=0.73 vs R=-0.12), cluster-density metric (CD=0.51 vs 0.68), human-panel validation on prospective novel odorants, linear projection readouts for strength/similarity/applicability

Confirmed in models

Principal Odor Map (Message Passing Neural Network) · Principal Odor Map GNN

Method

Representational Similarity Analysis (RSA)

Papers

A Principal Odor Map Unifies Diverse Tasks in Human Olfactory Perception — Lee, Brian K., Mayhew, Emily J., Sanchez-Lengeling, Benjamin, Wei, Jennifer N., Qian, Wesley W., Little, Kelsie, Andres, Matthew, Nguyen, Britney B., Moloy, Theresa, Parker, Jane K., Gerkin, Richard C., Mainland, Joel D., Wiltschko, Alexander B.2023