Chemical Similarity Search Using Property-Aware Neural Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing chemical information databases struggle to accurately identify structurally similar chemicals and capture nuanced functional relationships due to the limitations of rule-based algorithms and fingerprint methods, which are often slow and inefficient.
Innovation Solution
Utilizing deep learning models, specifically graph neural networks, to generate embeddings of chemical structures that incorporate both structural and property information, forming a vector database for efficient chemical similarity searches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based algorithms and fingerprint methods are used for chemical structure comparison, then the search process is simple to implement, but the accuracy of identifying structurally similar chemicals deteriorates
Solution Approach 1:
The patent replaces traditional rule-based algorithms and fingerprint methods with a graph neural network model. The GNN learns structural similarities through message passing and aggregation operations on molecular graphs, substituting the mechanical rule-based approach with a learned neural network approach that captures nuanced structural relationships more accurately.
Solution Approach 2:
The patent transforms chemical structures into continuous vector embeddings rather than discrete fingerprint bits. By representing molecules as high-dimensional vectors that capture structural information in a continuous space, the system enables more precise similarity measurement through vector distance metrics, improving upon the binary fingerprint comparison approach.
2Device complexity
If rule-based algorithms are used for chemical similarity search, then the computational complexity is low, but the efficiency and speed of search deteriorates
Solution Approach 1:
The patent pre-computes and stores embeddings for all chemicals in the database before query time. This preliminary action allows query chemicals to be quickly compared against pre-processed representations, significantly accelerating search speed while maintaining manageable computational complexity during actual queries.
Solution Approach 2:
The patent creates compressed vector representations (embeddings) of chemical structures that capture essential structural information in a condensed form. These embeddings serve as efficient copies that enable rapid similarity computation without requiring complex operations on the full molecular structures during query processing.
3Device complexity
If traditional fingerprint methods are used, then the data representation is simple, but the ability to capture nuanced functional relationships deteriorates
Solution Approach 1:
The patent transitions from low-dimensional binary fingerprint vectors to high-dimensional continuous embedding vectors. This dimensional expansion allows the representation to capture nuanced functional relationships and subtle structural variations that are lost in compressed binary representations, while still maintaining computational tractability.
Solution Approach 2:
The patent changes the parameter space from discrete binary values to continuous real-valued vectors. This parameter transformation enables the representation to encode graded structural similarities and functional relationships, capturing nuanced information that binary fingerprints cannot represent.
Data Source
AI summary
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.