Machine learning driven chemical compound replacement technology

Abstract

Techniques to suggest alternative chemical compounds that can be used to recreate or mimic a target flavor using artificial intelligence are disclosed. A neural network based model is trained on source chemical compounds and their corresponding flavors and odors. The neural network-based model learns compound embeddings of the source chemical compounds and a target chemical compound of a food item. From the compound embeddings, one or more chemical compounds that are closest to the target chemical compound may be determined by a distance metric. Each suggested chemical compound is an alternative that can be used to recreate functional features of the target chemical compound.

What is claimed is: 
     
         1 . A computer-implemented method of suggesting chemical compounds as substitutes for a chemical compound, comprising:
 obtaining a plurality of source compound embeddings for a plurality of source chemical compounds, wherein the plurality of source compound embeddings for the plurality of source chemical compounds is obtained by applying an artificial intelligence model to the plurality of source chemical compounds, wherein the artificial intelligence model is trained based on chemical structures and flavors of the plurality of source chemical compounds;   obtaining a target compound embedding for a target chemical compound, wherein the target compound embedding for the target chemical compound is obtained by applying the artificial intelligence model to the target chemical compound; and   using the target compound embedding for the target chemical compound and the plurality of source chemical compound embeddings for the plurality of source chemical compounds to identify one or more of the plurality of source chemical compounds as one or more alternative chemical compounds for the target chemical compound.   
     
     
         2 . The method of  claim 1 , wherein the artificial intelligence model is trained by:
 transforming each of the plurality of source chemical compounds into a graph that includes nodes representing atoms of a respective source chemical compound and edges representing bonds between the atoms of the respective source chemical compound;   applying a graph neural network on the graph;   processing an output from layers of the graph neural network through dense layers of a feed forward network.   
     
     
         3 . The method of  claim 2 , wherein processing the output from the layers of the graph neural network comprises:
 generating a representation of the graph;   matching the representation of the graph with at least the flavors of the plurality of source chemical compounds.   
     
     
         4 . The method of  claim 3 , wherein generating the representation of the graph comprises performing a computation on node representations of the nodes in the graph. 
     
     
         5 . The method of  claim 2 , wherein the plurality of source compound embeddings for the plurality of source chemical compounds and the target compound embedding for the target chemical compound are retrieved from the dense layers of the feed forward network. 
     
     
         6 . The method of  claim 1 , wherein the one or more alternative chemical compounds are determined using a distance metric. 
     
     
         7 . The method of  claim 1 , wherein the target chemical compound is identified from a list of chemical compounds, wherein the identification comprises:
 performing gas chromatography mass spectrometry (GCMS) analysis on a particular product that includes the target chemical compound;   generating, from the GCMS analysis, the list of chemical compounds, wherein each of the chemical compounds in the list of chemical compounds is a chemical compound present in the particular product.   
     
     
         8 . The method of  claim 1 , further comprising determining one or more plant-based ingredients containing at least one of the one or more alternative chemical compounds. 
     
     
         9 . One or more non-transitory computer-readable storage media storing one or more instructions programmed for suggesting chemical compounds as substitutes for a chemical compound, when executed by one or more computing devices, cause:
 obtaining a plurality of source compound embeddings for the plurality of source chemical compounds, wherein the plurality of source compound embeddings for the plurality of source chemical compounds is obtained by applying an artificial intelligence model to the plurality of source chemical compounds, wherein the artificial intelligence model is trained based on chemical structures and flavors of the plurality of source chemical compounds;   obtaining a target compound embedding for a target chemical compound, wherein the target compound embedding for the target chemical compound is obtained by applying the artificial intelligence model to the target chemical compound; and   using the target compound embedding for the target chemical compound and the plurality of source chemical compound embeddings for the plurality of source chemical compounds to identify one or more of the plurality of source chemical compounds as one or more alternative chemical compounds for the target chemical compound.   
     
     
         10 . The one or more non-transitory computer-readable storage media of  claim 9 , wherein the artificial intelligence model is trained by:
 transforming each of the plurality of source chemical compounds into a graph that includes nodes representing atoms of a respective source chemical compound and edges representing bonds between the atoms of the respective source chemical compound;   applying a graph neural network on the graph;   processing an output from layers of the graph neural network through dense layers of a feed forward network.   
     
     
         11 . The one or more non-transitory computer-readable storage media of  claim 10 , wherein processing the output from the layers of the graph neural network comprises:
 generating a representation of the graph;   matching the representation of the graph with at least the flavors of the plurality of source chemical compounds.   
     
     
         12 . The one or more non-transitory computer-readable storage media of  claim 11 , wherein generating the representation of the graph comprises performing a computation on node representations of the nodes in the graph. 
     
     
         13 . The one or more non-transitory computer-readable storage media of  claim 10 , wherein the plurality of source compound embeddings for the plurality of source chemical compounds and the target compound embedding for the target chemical compound are retrieved from the dense layers of the feed forward network. 
     
     
         14 . The one or more non-transitory computer-readable storage media of  claim 9 , wherein the one or more alternative chemical compounds are determined using a distance metric. 
     
     
         15 . The one or more non-transitory computer-readable storage media of  claim 9 , wherein the target chemical compound is identified from a list of chemical compounds, wherein the identification comprises:
 performing gas chromatography mass spectrometry (GCMS) analysis on a particular product that includes the target chemical compound;   generating, from the GCMS analysis, the list of chemical compounds, wherein each of the chemical compounds in the list of chemical compounds is a chemical compound present in the particular product.   
     
     
         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:
 obtaining a plurality of source compound embeddings for a plurality of source chemical compounds, wherein the plurality of source compound embeddings for the plurality of source chemical compounds is obtained by applying an artificial intelligence model to the plurality of source chemical compounds, wherein the artificial intelligence model is trained based on chemical structures and flavors of the plurality of source chemical compounds; 
 obtaining a target compound embedding for a target chemical compound, wherein the target compound embedding for the target chemical compound is obtained by applying the artificial intelligence model to the target chemical compound; and 
 using the target compound embedding for the target chemical compound and the plurality of source chemical compound embeddings for the plurality of source chemical compounds to identify one or more of the plurality of source chemical compounds as one or more alternative chemical compounds for the target chemical compound. 
   
     
     
         17 . The computing system of  claim 16 , wherein the artificial intelligence model is trained by:
 transforming each of the plurality of source chemical compounds into a graph that includes nodes representing atoms of a respective source chemical compound and edges representing bonds between the atoms of the respective source chemical compound;   applying a graph neural network on the graph;   processing an output from layers of the graph neural network through dense layers of a feed forward network.   
     
     
         18 . The computing system of  claim 17 , wherein processing the output from the layers of the graph neural network comprises:
 generating a representation of the graph;   matching the representation of the graph with at least the flavors of the plurality of source chemical compounds.   
     
     
         19 . The computing system of  claim 18 , wherein generating the representation of the graph comprises performing a computation on node representations of the nodes in the graph. 
     
     
         20 . The computing system of  claim 17 , wherein the plurality of source compound embeddings for the plurality of source chemical compounds and the target compound embedding for the target chemical compound are retrieved from the dense layers of the feed forward network.