Junction Tree Variational Autoencoder for Molecular Graph Generation

By Wengong Jin, Regina Barzilay, Tommi S. Jaakkola. 2018

1,678 citations recorded by the source corpus.

We seek to automate the design of molecules based on specific chemical properties. In computational terms, this task involves continuous embedding and generation of molecular graphs. Our primary contribution is the direct realization of molecular graphs, a task previously approached by generating linear SMILES strings instead of graphs. Our junction tree variational autoencoder generates molecular graphs in two phases, by first generating a tree-structured scaffold over chemical substructures, and then combining them into a molecule with a graph message passing network. This approach allows us to incrementally expand molecules while maintaining chemical validity at every step. We evaluate our model on multiple tasks ranging from molecular generation to optimization. Across these tasks, our model outperforms previous state-of-the-art baselines by a significant margin.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Materials Chemistry, Generative Models, Probabilistic & Bayesian Methods

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Theory and Mathematics

In Materials Chemistry

In Generative Models

In Probabilistic & Bayesian Methods

Citations in the corpus

Score-based Generative Modeling in Latent Space (2021, methodology): “LSGM would allow to model graph data more naturally using encoders and decoders tailored to graphs [43, 44, 45, 46].”
Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generation (2018, methodology): “Junction Tree VAE (JT-VAE) [16] is a state-of-the-art algorithm that combines graph representation and a VAE framework for generating molecular graphs, and uses Bayesian optimization over the learned latent space to search for molecules with optimized property scores.”
Learning Multimodal Graph-to-Graph Translation for Molecular Optimization (2018, methodology): “As baselines, we utilize state-of-the-art graph generation methods (Jin et al., 2018; You et al., 2018a) and MMPA (Dalke et al.”
Graphite: Iterative Generative Modeling of Graphs (2018, background): “Latent variable generative models have also been proposed for generating small molecular graphs (Jin et al., 2018; Samanta et al., 2018; Simonovsky & Komodakis, 2018).”
GeoDiff: a Geometric Diffusion Model for Molecular Conformation Generation (2022, background): “, 2015) to molecule generation (Jin et al., 2018; Shi et al., 2020), where typically a molecule is represented as an atom-bond graph.”
Permutation Invariant Graph Generation via Score-Based Generative Modeling (2020, background): “, 2018), and junction tree variational autoencoders (Jin et al., 2018).”
Learning Gradient Fields for Molecular Conformation Generation (2021, background): “…of molecules, where nodes represent atoms and edges represent bonds, have been widely applied to a variety of molecular modeling tasks, ranging from property prediction (Gilmer et al., 2017; Hu et al., 2020) to molecule generation (Jin et al., 2018; You et al., 2018; Shi et al., 2020).”
Can graph neural networks count substructures? (2020, background): “Meanwhile, graph substructures (also referred to by various names including graphlets, motifs, subgraphs and graph fragments) are well-studied and relevant for graph-related tasks in computational chemistry [12, 14, 26, 27, 28, 45], computational biology [32] and social network studies [25].”
MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, methodology): “This may lead to the generation of invalid string [5], [121], [125].”
A 3D Generative Model for Structure-Based Drug Design (2022, methodology): “The majority of graph-based models generate molecules in an auto-regressive fashion, i.e., adding atoms or fragments sequentially, which could be implemented based upon VAEs [13], normalizing ﬂows [27], reinforcement learning [35, 14], etc.”
Graph Representation Learning in Biomedicine and Healthcare (2021, neutral): “Perspective https://doi.org/10.1038/s41551-022-00942-x latent distribution (in particular, through variational autoencoders 93–95 or generative adversarial networks 96 ).”
Design Space for Graph Neural Networks (2020, background): “These architectures are then applied to a growing number of applications, such as social networks [39, 46], chemistry [13, 44], and biology [52].”
Diffusion Models: A Comprehensive Survey of Methods and Applications (2022, background): “Graph Neural Networks [85, 251, 266, 285] and corresponding representation learning [86] techniques have achieved great success [14, 231, 250, 258, 264, 288] in many areas, including modeling molecule graph in various tasks ranging from property prediction [59, 71] to molecule generation [105, 111, ”
All SMILES Variational Autoencoder (2019, methodology): “The distance between different SMILES strings of the same molecule can be much greater than that between SMILES strings from radically dissimilar molecules [27], as shown in Figure 8 of Appendix A.”
Generating 3D Molecular Structures Conditional on a Receptor Binding Site with Deep Generative Models (2020, background): “Moreover, VAEs together with molecular graph representations [30, 31], which more naturally capture the invariances present in chemical structures, were proposed and can generate novel molecules”