Junction Tree Variational Autoencoder for Molecular Graph Generation

ICML 2018  ·  Wengong Jin, Regina Barzilay, Tommi Jaakkola ·

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.

PDF Abstract ICML 2018 PDF ICML 2018 Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Molecular Graph Generation InterBioScreen JTVAE19 Validity 100.0 # 1
Molecular Graph Generation ZINC JT-VAE Reconstruction 76.7% # 1
Validty 100% # 1
QED Top-3 0.925, 0.911, 0.910 # 1
PlogP Top-3 5.30, 4.93, 4.49 # 1
function evaluations 2500 # 2
Uniqueness 100 # 1
Novelty 100 # 1
NUV 100 # 1

Methods


No methods listed for this paper. Add relevant methods here