# US 2021/0280276 A1 · Using machine learning for generating chemical product formulations

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0280276 A1
- Document ID: US20210280276A1
- Status: Published application
- Kind: A1
- Application number: 16/811,844
- Filed: 2020-03-06
- Published: 2021-09-09
- Earliest claimed: 2020-03-06
- Assignees: Accenture Global Solutions Limited
- Applicants: Accenture Global Solutions Limited
- Inventors: Vivek GUPTA; Shreeballav Sahoo; Amar Ratanlal Bafna
- Web page: https://patentgrep.com/patent/US20210280276A1

## CPC

- G16C 20/10: Analysis or design of chemical reactions, syntheses or processes
- G06F 3/0482: Interaction with lists of selectable items, e.g. menus
- G06F 18/2113: by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
- G06N 3/0499: Feedforward networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 20/00: Machine learning
- G16C 20/70: Machine learning, data mining or chemometrics
- G16C 20/80: Data visualisation
- G16C: COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A chemical product formulation system automatically generates seed formulae from historic experiments data for the synthesis of a chemical product. Independent and dependent features are identified from the historic experiments data and feature importance scores are calculated using a supervised machine learning (ML) model. The feature importance scores are used to build data structures from which analytical rules are extracted. The analytical rules are further processed to derive the seed formulae which are user-editable. The intermediate formulae generated via user edits of the seed formulae are further validated and approved in order to be used as the final formulae which are employed for the synthesis of the chemical product.

## Claims

### Claim 1

An chemical product formulation system comprising:
at least one processor;
a non-transitory processor-readable medium storing machine-readable instructions that cause the processor to:
access historical experiments data that includes data regarding experiments for synthesis of a chemical product, the historical experiments data including ingredients used for making the chemical product, proportions of the ingredients required to make the chemical product and specifications associated with each of the experiments;
identify features from the historical experiments data;
calculate a feature importance score of each of the features using a machine learning (ML) algorithm;
select a subset of the features based on the feature importance scores;
build an ML model that includes a classification tree that encodes data pertaining to the subset of the features;
extract rules for synthesizing the chemical product from the ML model;
generate seed formulae for the synthesis of the chemical product from the rules, wherein the seed formulae include at least ingredients and quantities of the ingredients to be used for the synthesis of the chemical product; and
provide one or more of the seed formulae that are validated and stored as final formulae to a production system for making the chemical product.

### Claim 2 (depends on 1)

The chemical product formulation system of claim 1, wherein the processor is to further:
preprocess the historical experiments data by improving data density and via data deduplication.

### Claim 3 (depends on 1)

The chemical product formulation system of claim 1, wherein the processor is to further:
identify target features and explanatory features from the features extracted from the historical experiments data.

### Claim 4 (depends on 1)

The chemical product formulation system of claim 1, wherein the ML algorithm includes decision trees, and to select the subset of the features the processor is to further:
obtain the feature importance scores by calculating Gini index for each node in each decision tree.

### Claim 5 (depends on 4)

The chemical product formulation system of claim 4, wherein to select the subset of the features, the processor is to further:
normalize the feature importance values.

### Claim 6 (depends on 4)

The chemical product formulation system of claim 4, wherein to calculate the Gini index the processor is to further:
build the ML model including a classification and regression tree (CART).

### Claim 7 (depends on 6)

The chemical product formulation system of claim 6, wherein to build the CART the processor is to further:
evaluate split in feature selection using the Gini index as a cost function for the generation of the CART.

### Claim 8 (depends on 1)

The chemical product formulation system of claim 1, wherein to extract the rules for synthesizing the chemical product the processor is to:
identify a root node and child nodes of a classification and regression tree (CART) encoding the subset of features.

### Claim 9 (depends on 8)

The chemical product formulation system of claim 8, wherein to extract the rules for synthesizing the chemical product the processor is to:
extract node attributes of the root node and each of the child nodes, wherein the node attributes include at least names of the subset of features, threshold values associated with the features and mathematical operators associated with the threshold values.

### Claim 10 (depends on 9)

The chemical product formulation system of claim 9, wherein to extract the rules for synthesizing the chemical product the processor is to:
group the node attributes on a target state, wherein the target state includes a quality criterion for the synthesis of the chemical product.

### Claim 11 (depends on 10)

The chemical product formulation system of claim 10, wherein to extract the rules for synthesizing the chemical product the processor is to:
store the node attributes as entries into a textual data file.

### Claim 12 (depends on 1)

The chemical product formulation system of claim 1, wherein to generate the seed formulae for the synthesis of the chemical product the processor is to:
identify from the rules, the ingredients and range thresholds for the quantities of the ingredients to be used for synthesizing the chemical product in accordance with a quality criterion.

### Claim 13 (depends on 12)

The chemical product formulation system of claim 12, wherein to generate the seed formulae for the synthesis of the chemical product the processor is to:
enable a display of the ingredients and the quantities of the ingredients via a formulation graphical user interface (GUI), wherein the GUI enables further customization and validation of the seed formulae; and
receive one or more intermediate formulae for the synthesis of the chemical product, the intermediate formulae include user selections of specific values for the quantities.

### Claim 14 (depends on 13)

The chemical product formulation system of claim 13, wherein to provide the seed formulae that are validated to the production system the processor is to:
provide the intermediate formulae to a supervisory user for validation; and
receive validated intermediate formulae as the final formulae.

### Claim 15 (depends on 13)

The chemical product formulation system of claim 13, wherein to provide the seed formulae that are validated to the production system the processor is to:
transmit at least one of the final formulae to the production system for the synthesis of the chemical product.

### Claim 16

A method of synthesizing a chemical product comprising:
preprocessing historic experiments data that includes ingredients used for synthesizing the chemical product in prior experiments, quantities of the ingredients used for the synthesis of the chemical product in the historic experiments, specifications of the chemical product that was synthesized and outcomes of the historic experiments;
receiving information regarding independent features and dependent features from the historic experiments data;
obtaining feature importance scores of the independent features and the dependent features using a trained machine learning (ML) model;
constructing data structures encoding the historic experiments data based on the feature importance scores;
extracting analytical rules from the data structures wherein the analytical rules include the ingredients and the quantities of ingredients to be used expressed in terms of mathematical operators and thresholds including minimum and maximum quantities of the ingredients;
generating seed formulae from the analytical rules, the seed formulae including the ingredients and the quantities of ingredients to be used for the synthesis of the chemical product; and
enabling a user to generate one or more intermediate formulae from the seed formulae via displaying the seed formula on a formulation graphical user interface (GUI) that enables the user to select specific quantities for each of the ingredients; and
providing final formulae that correspond to validated intermediate formulae to a production system.

### Claim 17 (depends on 16)

The method of claim 16, wherein the trained ML model includes a random forest model or a neural network.

### Claim 18 (depends on 16)

The method of claim 16, wherein each of the seed formulae includes the ingredients and an allowable range of quantities of the ingredients to be added.

### Claim 19

A non-transitory processor-readable storage medium comprising machine-readable instructions that cause a processor to:
access historical experiments data that includes data regarding experiments for synthesis of a chemical product, the historical experiments data including ingredients used for making the chemical product, proportions of the ingredients required to make the chemical product and specifications associated with each of the experiments;
extract features from the historical experiments data using a feature selection algorithm;
calculate a feature importance score of each of the features using a machine learning (ML) algorithm;
select a subset of the features based on the feature importance scores;
build an ML model that includes a classification tree that encodes data pertaining to the subset of the features;
extract rules for synthesizing the chemical product from the ML model;
generate seed formulae for the synthesis of the chemical product from the rules, wherein the seed formulae include at least ingredients and quantities of the ingredients to be used for the synthesis of the chemical product; and
provide one or more of the seed formulae that are validated to a production system for production of the chemical product.

### Claim 20 (depends on 19)

The non-transitory processor-readable storage medium of claim 19, further comprising instructions that cause the processor to:
enable a user to generate one or more intermediate formulae from the seed formulae via displaying the seed formulae on a formulation graphical user interface (GUI) that enables the user to select specific quantities for each of the ingredients.