How does JT-VAE compare to GraphVAE for molecule generation?

JT-VAE ensures chemical validity through junction tree decomposition, while GraphVAE often produces invalid structures. JT-VAE is better for reliability in drug discovery, but may be slower due to the decomposition step.

How to install JT-VAE with RDKit on Ubuntu?

Use conda as recommended in the README: install RDKit via conda, then set up Python 2.7 and PyTorch. Refer to `fast_molvae/README.md` for detailed steps on the accelerated version.

What datasets work best for training JT-VAE?

JT-VAE is typically trained on chemical datasets like ZINC or ChEMBL, as mentioned in the README's update. The newer hgraph2graph repository includes a pre-trained model on ChEMBL for better performance.

Can JT-VAE generate molecules for specific properties like solubility?

Yes, use the property-guided scripts in `bo/` for Bayesian optimization or `molopt/` for joint training. These allow optimization in the latent space to target properties like solubility.

Is JT-VAE suitable for commercial drug discovery projects?

Yes, it's designed for property optimization, but check licensing and consider the newer hgraph2graph version for better support and Python 3 compatibility, as advised in the README.

How long does it take to train JT-VAE on a GPU?

Training times vary with dataset size, but the accelerated version reduces time. For example, on the ZINC dataset, it might take several hours to a day on a modern GPU, as implied by the optimization efforts.

JTVAE

MITPython

A junction tree variational autoencoder for generating valid molecular graphs with desired chemical properties.

GitHub

What is JTVAE?

JT-VAE is a deep learning model for generating molecular graphs with desired chemical properties. It uses a junction tree decomposition to represent molecules as trees of chemically valid substructures, combined with a variational autoencoder to learn latent representations. This enables property-guided molecule generation for applications like drug discovery and materials design.

Target Audience

Researchers and practitioners in computational chemistry, drug discovery, and materials science who need to generate novel molecules with specific properties using machine learning.

Value Proposition

It generates chemically valid molecules by enforcing syntactic constraints through junction tree decomposition, offers property-guided optimization, and provides an accelerated implementation for efficient training and inference.

Overview

Junction Tree Variational Autoencoder for Molecular Graph Generation (ICML 2018)

Use Cases

Best For

Generating novel drug-like molecules with optimized properties

Related Projects

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub

556 stars197 forks0 contributors

Exploring chemical space for materials design

Molecular optimization using Bayesian optimization

Learning latent representations of molecular structures

Pre-training molecular language models on large chemical datasets

Research in graph-based generative models for chemistry

Not Ideal For

Production environments needing real-time molecule generation with low latency
Teams without expertise in computational chemistry and deep learning
Projects requiring Python 3.x and modern machine learning frameworks like TensorFlow 2
Applications demanding cross-platform support beyond Linux

Pros & Cons

Pros

Chemically Valid Generation

Uses junction tree decomposition to enforce syntactic validity, ensuring all generated molecules are chemically plausible based on substructure trees.

Property Optimization Support

Integrates Bayesian optimization and joint training with property predictors, enabling targeted generation of molecules with specific desired traits like drug-likeness.

Accelerated Implementation

Includes optimized code in `fast_jtnn/` and `fast_molvae/` directories, significantly speeding up training and inference compared to the original version.

Research Provenance

Based on a peer-reviewed ICML paper, providing a reproducible methodology with clear experimental scripts in directories like `bo/` and `molopt/`.

Cons

Outdated Python Version

Requires Python 2.7, which is no longer supported, complicating integration with modern libraries and tools, as noted in the README.

Complex Dependency Setup

Necessitates installation of RDKit and specific PyTorch versions, making initial configuration error-prone, especially for users unfamiliar with conda.

Deprecation Warning

The README explicitly recommends using a newer repository (hgraph2graph), indicating this version may be less maintained and potentially buggy for current use.

Frequently Asked Questions

Home

Computational Biology

DiffDock

Implementation of DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking

Stars1,511

Forks349

Last commit1 year ago

Molecular Transformer

Molecular Transformer is a neural machine translation model adapted for chemistry that predicts chemical reaction outcomes and retrosynthetic pathways. It translates between molecular representations (SMILES strings) to forecast how molecules react or how target molecules can be synthesized, accelerating discovery in organic chemistry and drug development. ## Key Features - **Retrosynthesis Prediction** — Predicts reactant molecules needed to synthesize a target product molecule. - **Uncertainty Calibration** — Provides confidence estimates for predictions, helping chemists assess reliability. - **SMILES Tokenization** — Uses custom tokenization of SMILES strings to treat molecules as sequences for transformer models. - **Data Augmentation** — Doubles training data by generating random equivalent SMILES representations via RDKit. - **Pre-trained Models** — Includes models trained on public datasets (USPTO_MIT, USPTO_STEREO) with mixed or separated reactant/reagent formats. ## Philosophy Molecular Transformer aims to make AI-assisted chemical reaction prediction accessible to organic chemists, with the goal of integrating these models into daily laboratory workflows to accelerate molecular discovery.

Stars424

Forks82

Last commit4 years ago

REINVENT

REINVENT is a reinforcement learning framework specifically designed for de novo drug design, enabling the generation of novel molecular structures with optimized properties. It addresses the challenge of discovering new chemical entities by combining generative models with property prediction to explore chemical space efficiently. ## Key Features - **Reinforcement Learning Pipeline** — Uses RL to optimize molecular structures toward desired chemical properties and biological activities - **De Novo Molecular Generation** — Creates entirely new molecular entities rather than modifying existing compounds - **Property Optimization** — Incorporates scoring functions to guide generation toward molecules with specific target properties - **Template-Based Execution** — Provides configurable JSON templates for different running modes and experiments - **TensorBoard Integration** — Enables real-time monitoring and visualization of training logs and progress ## Philosophy REINVENT applies reinforcement learning principles to drug discovery, treating molecular generation as an optimization problem where the agent learns to propose molecules that maximize desired chemical and biological properties.

Stars373

Forks113

Last commit11 months ago

TargetDiff

The official implementation of 3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction (ICLR 2023)

Stars338

Forks53

Last commit2 years ago

#molecular-generation

#graph-neural-networks

#bayesian-optimization

#rdkit

#drug-discovery

#computational-chemistry

Computational Biology122