Embedding of Molecular Structure Using Molecular Hypergraph Variational Autoencoder with Metric Learning

Abstract

Deep learning approaches are widely used to search molecular structures for a candidate drug/material. The basic approach in drug/material candidate structure discovery is to embed a relationship that holds between a molecular structure and the physical property into a low‐dimensional vector space (chemical space) and search for a candidate molecular structure in that space based on a desired physical property value. Deep learning simplifies the structure search by efficiently modeling the structure of the chemical space with greater detail and lower dimensions than the original input space. In our research, we propose an effective method for molecular embedding learning that combines variational autoencoders (VAEs) and metric learning using any physical property. Our method enables molecular structures and physical properties to be embedded locally and continuously into VAEs’ latent space while maintaining the consistency of the relationship between the structural features and the physical properties of molecules to yield better predictions.

Locally Optimizable Joint Embedding Framework to Design Nitrogen‐rich Molecules that are Similar but Improved

“…The evaluation measures show significant and noticeable improvement with joint embedding. Previous works have shown that joint training leads to an organization of the data according to the prediction tasks, [25,27–29,36] and the present results appear to provide further support for this behavior. This has the effect presently of providing the observed clustering and continuity features of the latent space described above.…”

“…A recent discussion of joint training noted that it may fail to organize latent space dimensions not correlated with the jointly trained property and may result in the unwanted clustering of molecules with dissimilar structural features [36] . We note that our manual feature vector contains structure‐based properties many of which are uncorrelated.…”

Recent Advances in Variational Autoencoders With Representation Learning for Biomedical Informatics: A Survey

“…Yan et al [30] propose a re-balancing VAE Loss to generate more valid SMILES molecules. [34] Generating graphs from a continuous representation in the latent space graph QM9 and ZINC Shervani-Tabar et al 2020 [35] Computing the statistics of molecular properties given small size training data set graph QM9 [38] A VAE-based masked graph model for molecular graphs graph QM9 and ChEMBL Koge et al 2020 [39] Molecular embedding learning that combines metric learning and VAEs graph QM9 Kwon et al 2020 [40] Generates compressed graph representation for scalable molecular graphs graph ChEMBL…”

“…In addition to generating new molecules, there are also studies on representation learning for decision-making based on VAEs. Koge et al [39] proposed a method of molecular embedding learning using a combination of VAE and metric learning. This method can simultaneously maintain the consistency of the relationship between molecular structural features and physical properties, resulting in better predictions.…”

“…In addition, a key benefit of VAEs is the ability to control the distribution of the latent representation vector z, which can combine VAEs with representation learning to further improve the downstream tasks [18,22]. Moreover, the generated image quality and diversity are improved by the existing VAE-variants such as β-VAE [23] and InfoVAE [24], which combine VAEs with disentanglement [25], [26], [27], [28], [29], [30], [31] graph representation design [32], [33], [34], [35], [36], [37], [38], [39], [40] sequence datasets analyses sequence engineering [41], [42], [43], [44], [45], [46], [47] dimensionality reduction [48], [49], [50], [51], [52], [53], [54], [55] integrated multi-omics data analyses [56], [57] predict effects of mutations [58], [47] gene expression analyses [56], [59], [54], [60], [61], [57]<...>…”