Granted Patent B2

US 12,205,682 · App. 17/835,795 · Granted Jan 21, 2025

Inventors: Hojin Kang (Santiago, CL); Kyohei Kaneko (Oakland, CA); Francisco Clavero (Santiago, CL); Aadit Patel (Burlingame, CA); Isadora Nun (San Francisco, CA); Karim Pichara (San Francisco, CA)

Assignee: Notco Delaware, LLC

G16C20/70G06F16/9024G16C20/30

Abstract

Claims (71)

1. A computer-implemented method of suggesting chemical compounds, comprising:

applying a machine learning model to at least one of a first plurality of chemical compounds and a plurality of true flavor profiles corresponding to the first plurality of chemical compounds, to generate at least one of a plurality of compound projected embeddings and a plurality of positive flavor projected embeddings,

wherein the machine learning model includes a first neural network and a second neural network and is trained by, for each chemical compound in a second plurality of chemical compounds, aligning a compound projected embedding of a respective chemical compound with a positive flavor projected embedding associated with the respective chemical compound in a joint vector space, and distancing the compound projected embedding from one or more negative flavor projected embeddings associated with the respective chemical compound in the joint vector space,

wherein the compound projected embedding of the respective chemical compound is from a projection layer of the first neural network, and each of the positive flavor projected embedding and the one or more negative flavor projected embedding is from a projection layer of the second neural network;

performing a search against at least one of the plurality of compound projected embeddings or the plurality of positive flavor projected embeddings to determine one or more chemical compounds from the second plurality of chemical compounds that satisfy a request for suggested chemical compounds.

2. The method of claim 1 , wherein the machine learning model is part of an artificial intelligence model that also comprises:

a graph model generating a specific compound graph embedding for a specific chemical compound that is input to the artificial intelligence model;

a word model generating a specific flavor word embedding for a specific flavor profile that is input to the artificial intelligence model.

3. The method of claim 1 , further comprising:

transforming the respective chemical compound into a graph that includes nodes and edges;

applying a graph model on the graph;

applying the first neural network on an output of the graph model to generate the compound projected embedding;

applying a word model on each of the true flavor profile and the one or more false flavor profiles;

applying the second neural network on outputs of the word model to generate the positive flavor projected embedding and the one or more negative flavor negative projected embeddings.

4. The method of claim 1 , wherein the request identifies a particular chemical compound, and the method further comprises:

transforming the particular chemical compound into a graph that includes nodes and edges;

applying a graph model on the graph;

applying the first neural network on an output of the graph model to generate a particular compound projected embedding;

wherein the search is based on the particular compound projected embedding.

5. The method of claim 1 , wherein the request identifies a particular flavor profile, and the method further comprises:

applying a word model on the particular flavor profile;

applying the second neural network on output of the word model to generate a particular flavor projected embedding;

wherein the search is based on the particular flavor projected embedding.

6. The method of claim 1 , wherein the first plurality of chemical compounds includes at least the second plurality of chemical compounds.

7. The method of claim 1 , wherein the search is a similarity search.

8. The method of claim 1 , further comprising determining one or more plant-based ingredients containing the one or more chemical compounds.

9. One or more non-transitory computer-readable storage media storing one or more instructions programmed for suggesting chemical compounds, when executed by one or more computing devices, cause:

applying a machine learning model to at least one of a first plurality of chemical compounds and a plurality of true flavor profiles corresponding to the first plurality of chemical compounds, to generate at least one of a plurality of compound projected embeddings and a plurality of positive flavor projected embeddings,

wherein the machine learning model includes a first neural network and a second neural network and is trained by, for each chemical compound in a second plurality of chemical compounds, aligning a compound projected embedding of a respective chemical compound with a positive flavor projected embedding associated with the respective chemical compound in a joint vector space, and distancing the compound projected embedding from one or more negative flavor projected embeddings associated with the respective chemical compound in the joint vector space,

wherein the compound projected embedding of the respective chemical compound is from a projection layer of the first neural network, and each of the positive flavor projected embedding and the one or more negative flavor projected embedding is from a projection layer of the second neural network;

performing a search against at least one of the plurality of compound projected embeddings or the plurality of positive flavor projected embeddings to determine one or more chemical compounds from the second plurality of chemical compounds that satisfy a request for suggested chemical compounds.

10. The one or more non-transitory computer-readable storage media of claim 9 , wherein the machine learning model is part of an artificial intelligence model that also comprises:

a graph model generating a specific compound graph embedding for a specific chemical compound that is input to the artificial intelligence model;

a word model generating a specific flavor word embedding for a specific flavor profile that is input to the artificial intelligence model.

11. The one or more non-transitory computer-readable storage media of claim 9 , wherein the one or more instructions, when executed by the one or more computing devices, further cause:

transforming the respective chemical compound into a graph that includes nodes and edges;

applying a graph model on the graph;

applying the first neural network on an output of the graph model to generate the compound projected embedding;

applying a word model on each of the true flavor profile and the one or more false flavor profiles;

applying the second neural network on outputs of the word model to generate the positive flavor projected embedding and the one or more negative flavor negative projected embeddings.

12. The one or more non-transitory computer-readable storage media of claim 9 , wherein the request identifies a particular chemical compound, and wherein the one or more instructions, when executed by the one or more computing devices, further cause:

transforming the particular chemical compound into a graph that includes nodes and edges;

applying a graph model on the graph;

applying the first neural network on an output of the graph model to generate a particular compound projected embedding;

wherein the search is based on the particular compound projected embedding.

13. The one or more non-transitory computer-readable storage media of claim 9 , wherein the request identifies a particular flavor profile, and wherein the one or more instructions, when executed by the one or more computing devices, further cause:

applying a word model on the particular flavor profile;

applying the second neural network on output of the word model to generate a particular flavor projected embedding;

wherein the search is based on the particular flavor projected embedding.

14. The one or more non-transitory computer-readable storage media of claim 9 , wherein the search is a similarity search.

15. The one or more non-transitory computer-readable storage media of claim 9 , wherein the one or more instructions, when executed by the one or more computing devices, further cause determining one or more plant-based ingredients containing the one or more chemical compounds.

16. A computing system comprising:

one or more computer systems comprising one or more hardware processors and storage media; and

instructions stored in the storage media and which, when executed by the computing system, cause the computing system to perform:

applying a machine learning model to at least one of a first plurality of chemical compounds and a plurality of true flavor profiles corresponding to the first plurality of chemical compounds, to generate at least one of a plurality of compound projected embeddings and a plurality of positive flavor projected embeddings,

wherein the machine learning model includes a first neural network and a second neural network and is trained by, for each chemical compound in a second plurality of chemical compounds, aligning a compound projected embedding of a respective chemical compound with a positive flavor projected embedding associated with the respective chemical compound in a joint vector space, and distancing the compound projected embedding from one or more negative flavor projected embeddings associated with the respective chemical compound in the joint vector space,

wherein the compound projected embedding of the respective chemical compound is from a projection layer of the first neural network, and each of the positive flavor projected embedding and the one or more negative flavor projected embedding is from a projection layer of the second neural network;

performing a search against at least one of the plurality of compound projected embeddings or the plurality of positive flavor projected embeddings to determine one or more chemical compounds from the second plurality of chemical compounds that satisfy a request for suggested chemical compounds.

17. The computing system of claim 16 , wherein the machine learning model is part of an artificial intelligence model that also comprises:

a graph model generating a specific compound graph embedding for a specific chemical compound that is input to the artificial intelligence model;

a word model generating a specific flavor word embedding for a specific flavor profile that is input to the artificial intelligence model.

18. The computing system of claim 17 , wherein the request identifies a particular chemical compound, and wherein the instructions, when executed by the computing system, further cause the computing system to perform:

transforming the particular chemical compound into a graph that includes nodes and edges;

applying a graph model on the graph;

applying a compound projector on an output of the graph model to generate a particular compound projected embedding;

wherein the search is based on the particular compound projected embedding.

19. The computing system of claim 17 , wherein the request identifies a particular flavor profile, and wherein the instructions, when executed by the computing system, further cause the computing system to perform:

applying a word model on the particular flavor profile;

applying a flavor projector on output of the word model to generate a particular flavor projected embedding;

wherein the search is based on the particular flavor projected embedding.

20. The computing system of claim 17 , wherein the instructions, when executed by the computing system, further cause the computing system to perform determining one or more plant-based ingredients containing the one or more chemical compounds.

Assignments (1)

Continuity (2)

References Cited (133)