Small-Molecule Compound Clustering With Tagged Sample Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional clustering methods for small molecule compounds suffer from low identification tag efficiency, slow identification tag speed, excessive computation demand, and high storage space requirements, limiting the number of identified compounds to less than 100,000.
Innovation Solution
A method involving sample segmentation into subsets with initial identification tags, obtaining sample legends, and using both initial and legend identification tags to enhance accuracy and reduce processing space, employing coarse-grained clustering and compound graph processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional clustering methods based on chemical informatics are used, then compound clustering can be performed, but identification tag efficiency is low and identification tag speed is slow
Solution Approach 1:
The patent replaces traditional chemical informatics methods with a graph neural network-based deep learning system. The GNN model automatically learns molecular features from compound graphs, substituting manual feature extraction and traditional clustering algorithms with an automated neural network approach that achieves both high speed and high efficiency in identification tagging.
Solution Approach 2:
The patent uses fingerprint features as simplified representations (copies) of complex molecular structures. By computing similarities based on these fingerprint copies rather than processing full molecular structures, the system achieves fast identification tagging while maintaining clustering accuracy.
2Measurement precision
If similarity of fingerprint features is used for identification, then identification accuracy can be improved, but computation demand and storage space requirements become excessive
Solution Approach 1:
The patent segments the compound dataset into multiple batches for processing. The GNN model processes compounds in batches rather than all at once, reducing peak computation demand while maintaining identification accuracy. This batched processing approach allows the system to handle large datasets without excessive memory requirements.
Solution Approach 2:
The patent changes the representation parameters from traditional fingerprint bit vectors to continuous embedding vectors learned by the GNN. This parameter transformation enables more efficient similarity computation and reduces storage requirements while improving identification accuracy through learned feature representations.
3Quantity of substance
If traditional clustering methods are used, then clustering can be performed, but the number of identified compounds is limited to less than 100,000
Solution Approach 1:
The patent performs preliminary pre-clustering to generate initial clusters before applying the GNN-based identification. This preliminary action organizes the large dataset into manageable groups, enabling the system to process and identify compounds beyond 100,000 while maintaining high throughput through efficient batch processing of pre-organized data.
Data Source
AI summary
The present application provides a method, a device, and a system for compound clustering and a storage medium, including acquiring samples of compounds to be identified, and segmenting the samples of the compounds into a subset of samples including an initial identification tag; obtaining a sample legend according to the sample subset; and obtaining target identification tag results corresponding to the samples of the compounds according to the sample legend and the identification tag, wherein the identification tag includes the initial identification tag. The present invention provides an efficient, rapid and accurate method for small molecule compound clustering based on statistical compound clustering, improving the accuracy of small molecule compound clustering, reducing the processing space of clustering and breaking through the limitations of small molecule clustering, thereby making the processing of small molecule compound clustering more efficient and accurate.