Predicting olfactory properties of molecules using machine learning

1 . A computer-based method of predicting data related to olfactory properties of a molecule characterized by a chemical structure, the method comprising:

by a computer:

upon obtaining data informative of a spatial surface representation (SSR) of the molecule corresponding to the chemical structure thereof, selecting on the SSR a plurality of N surface points;

for each selected surface point, obtaining local data informative of spatial location on the SSR and local physicochemical properties of the selected surface point, thus giving rise to a surface points representation (SPR);

inputting data informative of the SPR into a Machine-Learned (ML) model trained to provide, in accordance with SPR, prediction data related to at least one olfactory property; and

receiving, as an output of the ML model, prediction data related to the at least one olfactory property of the molecule.

2 . The method of claim 1 , further comprising using the prediction data to enable fabricating one or more products related to the molecule, wherein the fabricating comprises at least one of: generating a recipe for a desired odor, designing synthetic new mono-molecules and/or mixtures with desired olfactory features, reformulating a given recipe with no impact on the resulting odor, and synthesizing the molecule.

3 . The method of claim 1 , further comprising using the data informative of SPR as an input to the ML model further trained to recognize one or more SPR patches corresponding to one or more odor primaries; wherein the outputted prediction data related to the at least one olfactory property of the molecule are informative of the one or more odor primaries expected to be perceived by interaction with the molecule.

4 . The method of claim 3 , further comprising using the received prediction data informative of odor primaries for odor digitization.

5 . The method of claim 1 , further comprising pre-processing the SPR, wherein the pre-processing results are used as the data informative of the SPR, wherein the pre-processing of the SPR optionally comprises transforming the selected surface points into a two-dimensional matrix.

6 . One or more computing devices comprising processors and memory, the one or more computing devices configured, via computer-executable instructions, to perform operations for operating, in a cloud computing environment, a system capable of predicting data related to olfactory properties of a molecule characterized by a chemical structure, the operations comprising:

upon obtaining data informative of a spatial surface representation (SSR) of the molecule corresponding to the chemical structure thereof, selecting on the SSR a plurality of N surface points;

for each selected surface point, obtaining local data informative of spatial location on the SSR and local physicochemical properties of the selected surface point, thus giving rise to a surface points representation (SPR);

inputting data informative of the SPR into a Machine-Learned (ML) model trained to provide, in accordance with SPR, prediction data related to at least one olfactory property; and

receiving, as an output of the ML model, prediction data related to the at least one olfactory property of the molecule.

7 . A non-transitory computer-readable medium storing instructions that, when executed by a computing system comprising a memory storing a plurality of program components executable by the computing system, cause the computing system to perform:

upon obtaining data informative of a spatial surface representation (SSR) of a molecule corresponding to a chemical structure thereof, selecting on the SSR a plurality of N surface points;

for each selected surface point, obtaining local data informative of spatial location on the SSR and local physicochemical properties of the selected surface point, thus giving rise to a surface points representation (SPR);

inputting data informative of the SPR into a Machine-Learned (ML) model trained to provide, in accordance with SPR, prediction data related to at least one olfactory property; and

receiving, as an output of the ML model, prediction data related to the at least one olfactory property of the molecule.

8 . A computer-based method of predicting data related to olfactory properties of a molecule characterized by a chemical structure, the method comprising:

by a computer:

upon obtaining data informative of a spatial surface representation (SSR) of the molecule corresponding to the chemical structure thereof, selecting on the SSR a plurality of N surface points;

for each selected surface point, obtaining local data informative of spatial location on the SSR and local physicochemical properties of the selected surface point, thus giving rise to a surface points representation (SPR);

inputting data informative of the SPR into a Machine-Learned (ML) model trained to recognize one or more SPR patches corresponding to one or more odor primaries; and

receiving, as an output of the ML model, prediction data informative of a combination of odor primaries expected to be perceived by interaction with the molecule.

9 . The method of claim 8 , further comprising:

sending to an odor emission unit prediction data informative of the combination of odor primaries, the combination characterized by IDs of respective odor primaries and a proportion thereof, wherein the odor emission comprises a diffuser with a set of substances, each substance enabling smell perception of a respective odor primary; and

enabling the odor emission to emit substances from the set of substances in accordance with the received data on the combination of odor primaries expected to be perceived by interaction with the molecule.

10 . One or more computing devices comprising processors and memory, the one or more computing devices configured, via computer-executable instructions, to perform operations for operating, in a cloud computing environment, a system capable of predicting data related to olfactory properties of a molecule characterized by a chemical structure, the operations comprising:

upon obtaining data informative of a spatial surface representation (SSR) of the molecule corresponding to the chemical structure thereof, selecting on the SSR a plurality of N surface points;

for each selected surface point, obtaining local data informative of spatial location on the SSR and local physicochemical properties of the selected surface point, thus giving rise to a surface points representation (SPR);

inputting data informative of the SPR into a Machine-Learned (ML) model trained to recognize one or more SPR patches corresponding to one or more odor primaries; and

receiving, as an output of the ML model, prediction data informative of a combination of odor primaries expected to be perceived by interaction with the molecule.

11 . The one or more computing devices of claim 10 , wherein the operations further comprising:

sending to an odor emission unit prediction data informative of the combination of odor primaries, the combination characterized by IDs of respective odor primaries and a proportion thereof, wherein the odor emission comprises a diffuser with a set of substances, each substance enabling smell perception of a respective odor primary; and

enabling the odor emission to emit substances from the set of substances in accordance with the received data on the combination of odor primaries expected to be perceived by interaction with the molecule.

12 . A non-transitory computer-readable medium storing instructions that, when executed by a computing system comprising a memory storing a plurality of program components executable by the computing system, cause the computing system to perform:

upon obtaining data informative of a spatial surface representation (SSR) of a molecule corresponding to a chemical structure thereof, selecting on the SSR a plurality of N surface points;

for each selected surface point, obtaining local data informative of spatial location on the SSR and local physicochemical properties of the selected surface point, thus giving rise to a surface points representation (SPR);

inputting data informative of the SPR into a Machine-Learned (ML) model trained to recognize one or more SPR patches corresponding to one or more odor primaries; and

receiving, as an output of the ML model, prediction data informative of a combination of odor primaries expected to be perceived by interaction with the molecule.

13 . A computer-based method of predicting a molecular chemical structure that enables one or more olfactory properties, the method comprising:

by a computer:

upon receiving requirement data informative of requirements related to at least one olfactory property of a molecule, applying to the requirement data a machine-learned (ML) model trained to predict, in accordance with requirement data, surface points representation (SPR) informative of local physicochemical properties of a plurality of points located on a spatial surface representation (SSR) of the molecule; receiving, as an output of the ML model, data informative of a predicted SPR corresponding to the requirement data; and

using the predicted SPR to calculate a predicted SSR and a corresponding predicted chemical structure of a molecule that would match the requirements.

14 . The method of claim 13 , wherein the local physicochemical properties are characterized by one or more parameters selected from the group comprising: curvature, wave kernel signature, heat kernel signal, geometric and distance parameters, electronegativity, electron affinity, masses, partial charge, free electrons-protons, and hydrophobicity.

15 . The method of claim 13 , further comprising using the prediction data to enable fabricating one or more products related to the molecule, wherein the fabricating comprises at least one of: generating a recipe for a desired odor, designing synthetic new mono-molecules and/or mixtures with desired olfactory features, reformulating a given recipe with no impact on the resulting odor, and synthesizing the molecule.

16 . The method of claim 13 , wherein the requirement data specify an odor as a weighted combination of odor primaries and wherein the ML model is trained to predict a patch of a surface point representation (SPR) in accordance with the given odor primary, the method further comprising:

a) applying the trained ML model to each odor primary specified by the requirements data;

b) for each given odor primary, receiving, as an output of the ML model, data informative of a predicted patch of SPR; and

c) using the SPR patches predicted to the odor primaries in the requested weighted combination to calculate an SPR corresponding to the required odor, the SPR usable for calculating the predicted chemical structure of the molecule that would match the requirements.

17 . One or more computing devices comprising processors and memory, the one or more computing devices configured, via computer-executable instructions, to perform operations for operating, in a cloud computing environment, a system capable of predicting a molecular chemical structure that enables one or more olfactory properties, the operations comprising:

upon receiving requirement data informative of requirements related to at least one olfactory property of a molecule, applying to the requirement data a machine-learned (ML) model trained to predict, in accordance with requirement data, surface points representation (SPR) informative of local physicochemical properties of a plurality of points located on a spatial surface representation (SSR) of the molecule; receiving, as an output of the ML model, data informative of a predicted SPR corresponding to the requirement data; and

using the predicted SPR to calculate a predicted SSR and a corresponding predicted chemical structure of a molecule that would match the requirements.

18 . A non-transitory computer-readable medium storing instructions that, when executed by a computing system comprising a memory storing a plurality of program components executable by the computing system, cause the computing system to perform:

upon receiving requirement data informative of requirements related to at least one olfactory property of a molecule, applying to the requirement data a machine-learned (ML) model trained to predict, in accordance with requirement data, surface points representation (SPR) informative of local physicochemical properties of a plurality of points located on a spatial surface representation (SSR) of the molecule;

receiving, as an output of the ML model, data informative of a predicted SPR corresponding to the requirement data; and

using the predicted SPR to calculate a predicted SSR and a corresponding predicted chemical structure of a molecule that would match the requirements.