Using machine learning to explore formulations recipes with new ingredients

A non-transitory computer-readable medium comprising memory with instructions encoded thereon, the instructions, when executed by one or more processors, causing operations the one or more processors to perform operations, the instructions comprising instructions to:

• receive first input from a user of a candidate formulation recipe;
• receive second input from the user of one or more target properties and target property constraints;
• input the first input and the second input into a machine learning model, the machine learning model having been trained using historical training data, each element of the historical training data corresponding to a known formulation having a known feature representation, each known formulation having associated properties and statistical representations of each feature of the known formulation that form the known feature representation;
• receive as output from the machine learning model a predicted property of a candidate formulation, and a likelihood that the predicted property satisfies a constraint using the second input;
• and
• output for display to the user a predicted likelihood that the predicted property satisfies the second input.

A system comprising:

• memory with instructions encoded thereon;
• and
• one or more processors that, when executing the instructions, are caused to perform operations comprising:
• receiving first input from a user of a candidate formulation recipe;
• receiving second input from the user of one or more target properties and target property constraints;
• inputting the first input and the second input into a machine learning model, the machine learning model having been trained using historical training data, each element of the historical training data corresponding to a known formulation having a known feature representation, each known formulation having associated properties and statistical representations of each feature of the known formulation that form the known feature representation;
• receiving as output from the machine learning model a predicted property of a candidate formulation, and a likelihood that the predicted property satisfies a constraint using the second input;
• and
• outputting for display to the user a predicted likelihood that the predicted property satisfies the second input.