Search

Background: Converting 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. Results: In 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. Conclusions: This 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. [ABSTRACT FROM AUTHOR]

Motivation: Chemical–genetic interaction profiling is a genetic approach that quantifies the susceptibility of a set of mutants depleted in specific gene product(s) to a set of chemical compounds. With the recent advances in artificial intelligence, chemical–genetic interaction profiles (CGIPs) can be leveraged to predict mechanism of action of compounds. This can be achieved by using machine learning, where the data from a CGIP is fed into the machine learning platform along with the chemical descriptors to develop a chemogenetically trained model. As small molecules can be considered non-structural data, graph convolutional neural networks, which can learn from the chemical structures directly, can be used to successfully predict molecular properties. Clustering analysis, on the other hand, is a critical approach to get insights into the underlying biological relationships between the gene products in the high-dimensional chemical-genetic data. Methods and results: In this study, we proposed a comprehensive framework based on the large-scale chemical-genetics dataset built in Mycobacterium tuberculosis for predicting CGIPs using graph-based deep learning models. Our approach is structured into three parts. First, by matching M. tuberculosis genes with homologous genes in Escherichia coli (E. coli) according to their gene products, we grouped the genes into clusters with distinct biological functions. Second, we employed a directed message passing neural network to predict growth inhibition against M. tuberculosis gene clusters using a collection of 50,000 chemicals with the profile. We compared the performance of different baseline models and implemented multi-label tasks in binary classification frameworks. Lastly, we applied the trained model to an externally curated drug set that had experimental results against M. tuberculosis genes to examine the effectiveness of our method. Overall, we demonstrate that our approach effectively created M. tuberculosis gene clusters, and the trained classifier is able to predict activity against essential M. tuberculosis targets with high accuracy. Conclusion: This work provides an analytical framework for modeling large-scale chemical-genetic datasets for predicting CGIPs and generating hypothesis about mechanism of action of novel drugs. In addition, this work highlights the importance of graph-based deep neural networks in drug discovery. [ABSTRACT FROM AUTHOR]