Application of artificial intelligence to decode the relationships between smell, olfactory receptors and small molecules

Abstract

Deciphering the relationship between molecules, olfactory receptors (ORs) and corresponding odors remains a challenging task. It requires a comprehensive identification of ORs responding to a given odorant. With the recent advances in artificial intelligence and the growing research in decoding the human olfactory perception from chemical features of odorant molecules, the applications of advanced machine learning have been revived. In this study, Convolutional Neural Network (CNN) and Graphical Convolutional Network (GCN) models have been developed on odorant molecules-odors and odorant molecules-olfactory receptors using a large set of 5955 molecules, 160 odors and 106 olfactory receptors. The performance of such models is promising with a Precision/Recall Area Under Curve of 0.66 for the odorant-odor and 0.91 for the odorant-olfactory receptor GCN models respectively. Furthermore, based on the correspondence of odors and ORs associated for a set of 389 compounds, an odor-olfactory receptor pairwise score was computed for each odor-OR combination allowing to suggest a combinatorial relationship between olfactory receptors and odors. Overall, this analysis demonstrate that artificial intelligence may pave the way in the identification of the smell perception and the full repertoire of receptors for a given odorant molecule.

Figure 1

(A) Occurrence of compounds related to odors. (B) Occurrence of compounds related to olfactory receptors.

Figure 2

UMAP representation of compounds distribution in a 2D map projection. (A) Compounds with green odor note, (B) Compounds with woody odor note, (C) Compounds with spicy odor note, (D) Compounds with fruity odor note. The UMAP representations were developed with the python package bokeh (v.2.4.3): https://docs.bokeh.org/e.

Figure 3

UMAP representation of compounds distribution in a 2D map projection. (A) Compounds active on OR1D2, (B) Compounds active on OR5D16, (C) Compounds active on OR1A1, (D) Compounds active on OR2D4. The UMAP representations were developed with the python package bokeh (v.2.4.3): https://docs.bokeh.org/e.

Figure 4

Radar plot based on the frequency of physicochemical properties observed on the set compounds associated to an odor note. (A) Acidic, (B) Citrus, (C) Cheese, (D) Apple, (E) Muguet, (F) Floral. The radarplots were developed with the python package plotly (v.5.3.1): https://plotly.com/.

Figure 5

Radar plot based on the frequency of physicochemical properties observed on the set compounds associated to an olfactory receptor. (A) OR52E1, (B) OR4D6, (C) OR1D3, (D) OR1G1. The radarplots were developed with the python package plotly (v.5.3.1): https://plotly.com/.

Figure 6

Heatmap representation of the performance with the GCN model for a subset of the compound-odor note associations. The odor notes are represented on the X-axis and the compounds on the Y axis. A compound associated to an odor note and correctly classified by the GCN model is colored with a dark green cell. A compound not associated to an odor note and correctly predicted has a light green color. A compound linked to an odor note and wrongly predicted by the model is represented with a pink color and a compound not linked to an odor note and predicted to be associated to this odor note is shown with a red color. The heatmap was developed with the python package seaborn (v.0.11.2): https://seaborn.pydata.org/^,.

Figure 7

Heatmap representation of the performance prediction with the GCN model for a subset of the compound-olfactory receptors (ORs) associations. The ORs are represented on the X-axis and the compounds on the Y axis. A compound associated to an OR and correctly classified by the GCN model is colored with a dark green cell. A compound not associated to an OR and correctly predicted has a light green color. A compound linked to an OR and wrongly predicted by the model is represented in a pink color and a compound not linked to an OR and predicted to be associated to this odor is shown with a red color. The heatmap was developed with the python package seaborn (v.0.11.2): https://seaborn.pydata.org/^,.

Figure 8

Heatmap representation of the odor note-olfactory receptor pairwise score for a set of 383 compounds targeting olfactory receptors and related to odor notes. The ORs are represented on the X-axis and the odor notes on the Y-axis. A red dot represents a high odor note-olfactory receptor pairwise score and a dark blue dot, no odor note-olfactory receptor relationship. The heatmap was developed with the python package seaborn (v.0.11.2): https://seaborn.pydata.org/^–.