# US 2026/0066062 A1 · Chemical similarity search using fine-tuned neural network

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

- Number: US 2026/0066062 A1
- Document ID: US20260066062A1
- Status: Published application
- Kind: A1
- Application number: 18/825,994
- Filed: 2024-09-05
- Published: 2026-03-05
- Earliest claimed: 2024-09-05
- Assignees: Microsoft Technology Licensing, LLC
- Applicants: Microsoft Technology Licensing, LLC
- Inventors: Chi CHEN; Hongbin LIU; Yousif Waleed ALMULLA
- Web page: https://patentgrep.com/patent/US20260066062A1

## CPC

- G16C 20/40: Searching chemical structures or physicochemical data
- G16C 20/90: Programming languages; Computing architectures; Database systems; Data warehousing
- G16C 20/30: Prediction of properties of chemical compounds, compositions or mixtures
- G16C 20/70: Machine learning, data mining or chemometrics
- G16C: COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE

## Abstract

Examples are disclosed that relate to forming embeddings comprising vector representations of chemical structures, and performing chemical similarity searches. One example provides a method of forming a vector database using a pre-trained neural network, the method comprising inputting labeled training data into the pre-trained neural network configured to form embeddings of chemical structures, the labeled training data comprising structural information and a value of a property for each chemical object in a first set of chemical objects. The method further comprises fine-tuning the pre-trained neural network and forming the vector database by inputting a reference dataset into the fine-tuned neural network to generate embeddings of chemical structures, the reference dataset comprising structural information for each chemical object in a second set of chemical objects. Each embedding stored in the vector database comprises a vector representation of a chemical structure and embedded information for the property for the chemical structure.

## Claims

### Claim 1

A method of forming a vector database using a fine-tuned neural network, the method comprising:
inputting labeled training data into a pre-trained neural network configured to form embeddings of chemical structures, the labeled training data comprising structural information and a value of a property for each chemical object in a first set of chemical objects;
fine-tuning the pre-trained neural network using the labeled training data, thereby forming the fine-tuned neural network; and
forming the vector database by
inputting a reference dataset into the fine-tuned neural network to generate embeddings of chemical structures, the reference dataset comprising structural information for each chemical object in a second set of chemical objects, each embedding comprising a vector representation of a chemical structure and embedded information for the property for the chemical structure, and
storing the embeddings with metadata in the vector database.

### Claim 2 (depends on 1)

The method of claim 1, further comprising,
prior to inputting the labeled training data, inputting a set of pre-training data into a neural network, the set of pre-training data comprising structural information for each chemical object of a third set of chemical objects; and
training the neural network to form the pre-trained neural network.

### Claim 3 (depends on 1)

The method of claim 1, further comprising:
receiving a query comprising chemical structure information for a target chemical object;
based on the query, inputting the chemical structure information for the target chemical object into the fine-tuned neural network;
receiving an embedding for the target chemical object from the fine-tuned neural network, the embedding comprising a vector representation of the target chemical object;
based at least on a similarity score between the embedding for the target chemical object and each of one or more embeddings stored in the vector database, retrieving query results from the vector database, the query results comprising a set of embeddings and metadata for a corresponding set of chemical objects; and
outputting the query results.

### Claim 4 (depends on 3)

The method of claim 3, wherein outputting the query results comprises, for each chemical object in the corresponding set of chemical objects of the query results, outputting structural information and a predicted property value for the chemical object.

### Claim 5 (depends on 3)

The method of claim 3, wherein retrieving the query results from the vector database comprises determining one or more of a Euclidean distance, a Manhattan distance, a dot product, or a cosine similarity for the embedding for the target chemical object and an embedding stored in the vector database.

### Claim 6 (depends on 1)

The method of claim 1, wherein the value for the property comprises one or more of a scalar value or a classification for the property for each chemical object in the first set of chemical objects.

### Claim 7 (depends on 1)

The method of claim 1, wherein the labeled training data comprises values for two of more properties for each chemical object in the first set of chemical objects.

### Claim 8 (depends on 1)

The method of claim 1, wherein fine-tuning the neural network comprises fine-tuning a graph neural network (GNN).

### Claim 9

A method enacted on a computing system, the method comprising:
inputting labeled training data into a pre-trained neural network configured to form embeddings of chemical structures, the labeled training data comprising structural information and a value for a property for each chemical object in a set of chemical objects;
fine-tuning the pre-trained neural network using the labeled training data, thereby forming a fine-tuned neural network;
forming a vector database by using the fine-tuned neural network to generate embeddings of chemical structures of a reference dataset, each embedding comprising a vector representation of a chemical structure and embedded information for the property for the chemical structure, and storing the embeddings with metadata in the vector database;
receiving a query comprising chemical structure information for a target chemical object;
based on the query, inputting the chemical structure information into the fine-tuned neural network;
receiving an embedding for the target chemical object from the fine-tuned neural network, the embedding for the target chemical object comprising a vector representation of the target chemical object;
based at least on a similarity score between the embedding for the target chemical object and each of one or more embeddings stored in the vector database, retrieving query results from the vector database, the query results comprising a set of embeddings and metadata for a corresponding set of chemical objects; and
outputting the query results.

### Claim 10 (depends on 9)

The method of claim 9, further comprising,
prior to inputting the labeled training data, inputting a set of pre-training data into a neural network, the set of pre-training data comprising structural information for each chemical object of the dataset; and
training the neural network to form the pre-trained neural network.

### Claim 11 (depends on 9)

The method of claim 9, wherein the value for the property comprises one or more of a scalar value or a classification for the property for each chemical object in the set of chemical objects.

### Claim 12 (depends on 9)

The method of claim 9, wherein the labeled training data comprises values for two of more properties for each chemical object in the set of chemical objects.

### Claim 13 (depends on 9)

The method of claim 9, further comprising, for each chemical object in the corresponding set of chemical objects of the query results, outputting structural information and a predicted property value for the chemical object.

### Claim 14 (depends on 9)

The method of claim 9, wherein retrieving the query results from the vector database comprises determining one or more of a Euclidean distance, a Manhattan distance, a dot product, or a cosine similarity for the embedding for the target chemical object and an embedding stored in the vector database.

### Claim 15 (depends on 9)

The method of claim 9, wherein fine-tuning the neural network comprises fine-tuning a graph neural network (GNN).

### Claim 16

A computing system implementing a neural network configured to produce embeddings of chemical structures, the computing system comprising:
a logic subsystem; and
a storage subsystem comprising instructions executable by the logic subsystem to
input labeled training data into a pre-trained neural network configured to form embeddings of chemical structures, the labeled training data comprising structural information and a value for a property for each chemical object in a first set of chemical objects,
train the neural network to predict the property, thereby forming a fine-tuned neural network,
input a reference dataset into the fine-tuned neural network to generate embeddings of chemical structures, the reference dataset comprising structural information for a second set of chemical objects, each embedding comprising a vector representation of a chemical structure and embedded information for the property for the chemical structure, and
store the embeddings with metadata in a vector database.

### Claim 17 (depends on 16)

The computing system of claim 16, wherein the instructions are further executable to:
receive a query comprising chemical structure information for a target chemical object,
based on the query, input the chemical structure information into the fine-tuned neural network,
receive, from the fine-tuned neural network, an embedding comprising a vector representation of the target chemical object,
based at least on a similarity score between the embedding for the target chemical object and each of one or more embeddings stored in the vector database, retrieve query results from the vector database, the query results comprising a set of embeddings and metadata for a corresponding set of chemical objects, and
output the query results.

### Claim 18 (depends on 17)

The computing system of claim 17, wherein the instructions executable to output the query results are executable to output structural information and predicted property values for each chemical object in the corresponding set of chemical objects.

### Claim 19 (depends on 16)

The computing system of claim 16, wherein the labeled training data comprises values for two or more properties for each chemical object in the first set of chemical objects, and the instructions are executable to train the neural network to predict the two or more properties.

### Claim 20 (depends on 16)

The computing system of claim 16, wherein the instructions executable to train the neural network to predict the property comprise instructions executable to train a graph neural network to predict one or more of a boiling point, a flash point, an electronegativity, an ionization energy, a bandgap, or a dielectric constant.
