Deep clustering of small molecules at large-scale via variational autoencoder embedding and K-means

BMC Bioinformatics

Research

Rebecca Davis¹  Hamid Hadipour²  Pingzhao Hu³  Chengyou Liu⁴  Silvia T. Cardona⁵ 

[1] Department of Chemistry, University of Manitoba, Winnipeg, MB, Canada;Department of Computer Science, University of Manitoba, Winnipeg, MB, Canada;Department of Computer Science, University of Manitoba, Winnipeg, MB, Canada;Department of Electrical and Computer Engineering, University of Manitoba, Winnipeg, MB, Canada;Department of Biochemistry and Medical Genetics, University of Manitoba, Room 308 - Basic Medical Sciences Building, 745 Bannatyne Avenue, R3E 0J9, Winnipeg, MB, Canada;CancerCaree Manitoba Research Institute, CancerCare Manitoba, Winnipeg, MB, Canada;Department of Electrical and Computer Engineering, University of Manitoba, Winnipeg, MB, Canada;Department of Microbiology, University of Manitoba, Winnipeg, MB, Canada;Department of Medical Microbiology and Infectious Diseases, University of Manitoba, Winnipeg, MB, Canada;

关键词: Unsupervised deep clustering;    K-means;    Embedding;    Variational autoencoders;    Internal clustering measurements;    Chemical diversity;   

DOI  :  10.1186/s12859-022-04667-1

 received in 2022-03-23, accepted in 2022-04-04,  发布年份 2022

来源: Springer

【 摘 要 】

BackgroundConverting molecules into computer-interpretable features with rich molecular information is a core problem of data-driven machine learning applications in chemical and drug-related tasks. Generally speaking, there are global and local features to represent a given molecule. As most algorithms have been developed based on one type of feature, a remaining bottleneck is to combine both feature sets for advanced molecule-based machine learning analysis. Here, we explored a novel analytical framework to make embeddings of the molecular features and apply them in the clustering of a large number of small molecules.ResultsIn this novel framework, we first introduced a principal component analysis method encoding the molecule-specific atom and bond information. We then used a variational autoencoder (AE)-based method to make embeddings of the global chemical properties and the local atom and bond features. Next, using the embeddings from the encoded local and global features, we implemented and compared several unsupervised clustering algorithms to group the molecule-specific embeddings. The number of clusters was treated as a hyper-parameter and determined by the Silhouette method. Finally, we evaluated the corresponding results using three internal indices. Applying the analysis framework to a large chemical library of more than 47,000 molecules, we successfully identified 50 molecular clusters using the K-means method with 32 embeddings based on the AE method. We visualized the clustering result via t-SNE for the overall distribution of molecules and the similarity maps for the structural analysis of randomly selected cluster-specific molecules.ConclusionsThis study developed a novel analytical framework that comprises a feature engineering scheme for molecule-specific atomic and bonding features and a deep learning-based embedding strategy for different molecular features. By applying the identified embeddings, we show their usefulness for clustering a large molecule dataset. Our novel analytic algorithms can be applied to any virtual library of chemical compounds with diverse molecular structures. Hence, these tools have the potential of optimizing drug discovery, as they can decrease the number of compounds to be screened in any drug screening campaign.

【 授权许可】

CC BY   
© The Author(s) 2022

【 图 表 】

Fig. 1

Fig. 7

Fig. 3

Fig. 1

Fig. 3

41512_2023_153_Article_IEq102.gif

Fig. 2

Fig. 3

Fig. 1

Fig. 2

Fig. 1

Fig. 4

Fig. 4

13570_2023_282_Article_IEq21.gif

13570_2023_282_Article_IEq18.gif

13570_2023_282_Article_IEq15.gif