GraphAF: a Flow-based Autoregressive Model for Molecular Graph Generation

By Chence Shi, Minkai Xu, Zhaocheng Zhu, Weinan Zhang, Ming Zhang, Jian Tang. International Conference on Learning Representations, 2020

553 citations recorded by the source corpus.

Molecular graph generation is a fundamental problem for drug discovery and has been attracting growing attention. The problem is challenging since it requires not only generating chemically valid molecular structures but also optimizing their chemical properties in the meantime. Inspired by the recent progress in deep generative models, in this paper we propose a flow-based autoregressive model for graph generation called GraphAF. GraphAF combines the advantages of both autoregressive and flow-based approaches and enjoys: (1) high model flexibility for data density estimation; (2) efficient parallel computation for training; (3) an iterative sampling process, which allows leveraging chemical domain knowledge for valency checking. Experimental results show that GraphAF is able to generate 68% chemically valid molecules even without chemical knowledge rules and 100% valid molecules with chemical rules. The training process of GraphAF is two times faster than the existing state-of-the-art approach GCPN. After fine-tuning the model for goal-directed property optimization with reinforcement learning, GraphAF achieves state-of-the-art performance on both chemical property optimization and constrained property optimization.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology

View Paper: https://www.semanticscholar.org/paper/036d743c7ca1e513adf0a91594fc8111e03dc30c

How researchers see this paper

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

In Computational Theory and Mathematics

In Materials Chemistry

In Molecular Biology

Citations in the corpus

3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction (2023, methodology): “…and Threlfall, 2017; Kusner et al., 2017; Segler et al., 2018) or graph representation (Li et al., 2018; Liu et al., 2018; Jin et al., 2018; Shi et al., 2020), but both representations do not take the 3D spatial interaction into account and therefore not well suited for target-aware…”
TorchDrug: A Powerful and Flexible Machine Learning Platform for Drug Discovery (2022, background): “…et al., 2021), molecular property prediction (Duvenaud et al., 2015; Hu et al., 2019), de novo molecular design and optimization (You et al., 2018; Shi et al., 2020b) , reaction prediction (Jin et al., 2017; Bradshaw et al., 2018), retrosynthesis prediction (Dai et al., 2019; Shi et al., 2020a),…”
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).”
A 3D Generative Model for Structure-Based Drug Design (2022, background): “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.”
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.”
Score-based Generative Modeling of Graphs via the System of Stochastic Differential Equations (2022, methodology): “Following the standard procedure (Shi et al., 2020; Luo et al., 2021), the molecules are kekulized by the RDKit library (Landrum et al., 2016) and hydrogen atoms are removed.”
Iterative Deep Graph Learning for Graph Neural Networks: Better and Robust Node Embeddings (2020, background): “Other related works include structure inference in probabilistic graphical models [9, 60, 57], and graph generation [36, 46], which have a different goal from ours.”
Predicting Molecular Conformation via Dynamic Graph Score Matching (2021, background): “Graph-based representations of molecules has become prevalent in a variety of tasks such as property prediction [13, 31] and molecule generation [17, 33, 45].”
Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models (2018, methodology): “Explicit models such as Hidden Markov Models, n-gram language models, or normalizing flows (Dinh et al., 2017; Shi et al., 2019) can analytically compute q(x) and sample from it.”
Flow Network based Generative Models for Non-Iterative Diverse Candidate Generation (2021, methodology): “This sampling problem comes up in molecule generation and has been studied in this context with numerous generative models (Shi et al., 2020; Jin et al., 2020; Luo et al., 2021), MCMC methods (Seff et al., 2019; Xie et al., 2021), RL (Segler et al., 2017; Cao and Kipf, 2018; Popova et al., 2019;…”
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, ”
MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, unknown): “GraphAF [131] instead designs an autoregressive flow-based model to generate a 2D molecular graph sequentially, which adds an atom or a bond each step and guarantees validity via incorporating a valency check in each step.”
Differentiable Scaffolding Tree for Molecular Optimization (2021, unknown): “Typical algorithms include variational autoencoder (VAE), generative adversarial network (GAN), energy-based models, flow-based model [14, 21, 9, 37, 23, 17, 34, 38, 22, 49, 27, 5, 32, 1].”
Crystal Diffusion Variational Autoencoder for Periodic Material Generation (2021, methodology): “Flow models have also been used for molecular graph generation (Shi et al., 2020; Luo et al., 2021).”
MoFlow: An Invertible Flow Model for Generating Molecular Graphs (2020, methodology): “68 100 GraphAF [30] 100 67 94.”
Learning Neural Generative Dynamics for Molecular Conformation Generation (2021, background): “Recently, we have witnessed the success of graph-based representations for molecular modeling in a variety of tasks such as property prediction (Gilmer et al., 2017) and molecule generation (You et al., 2018; Shi et al., 2020).”
Continuous Graph Flow (2019, background): “Recent research attempts on normalizing flow based graph generative models (Liu et al., 2019; Madhawa et al., 2019; Shi et al., 2020; Honda et al., 2019) propose a reversible graph neural network using normalizing flows.”