OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Propmolflow

PropMolFlow is an open-source Python research model that generates molecular structures guided by desired properties using geometry-complete flow matching and SE(3)-equivariant representations.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Propmolflow — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use PropMolFlow when a research workflow needs property-conditioned 3D molecule generation and can accommodate the documented GPU-oriented Python environment and QM9 data-preparation steps.

Who it is for

  • Computational chemistry researchers exploring property-guided molecular generation.
  • Research teams preparing QM9-derived data to train or evaluate molecular generation models.
  • Practitioners comparing property-conditioning methods and pretrained model checkpoints.

Documented uses

  • Generate molecular candidates conditioned on target properties.
  • Prepare QM9-derived molecular structures and property data for training.
  • Train the model from scratch or continue training from a checkpoint.
  • Review benchmark results across property-handling and Gaussian-expansion configurations.
  • Calculate chemical validity and atom- and molecule-level valency-stability metrics from an SDF file.
Before you choose

Requirements & limits

The public GitHub repository specifies an MIT license. The documented local setup requires Conda and CUDA 12.8 or newer.

  • The repository provides 12 checkpoints for six properties across in-distribution and out-of-distribution tasks, and also describes a complete set of 180 checkpoints.
  • The model workflow is tied to QM9-derived SDF data and property-value CSV files, with documented problematic-index handling.
  • The installation instructions list CUDA-dependent packages and a GPU-based environment; the README reports installation testing on Nvidia L4 and Blackwell B200 GPUs.
  • A project demo website is referenced, while preprocessing and training instructions are provided in the repository.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Start training or resume from a checkpoint

Use the repository's documented training entry point with a configuration or existing checkpoint.

Before you start

  • A configured PropMolFlow environment.
  • Preprocessed QM9 data and a training configuration with the correct raw data paths and filenames.

Steps

  1. For training from scratch, run the command below.

    Documented command
    python train.py --config=configs/without_gaussian_expansion/alpha_sum.yaml

  2. To continue training, provide a checkpoint with the documented --resume argument and a configuration file.

  3. Set output_dir in the configuration if you want to change the output location.

Expected outcome

The training script saves model checkpoints in the configured output directory; the documented default is `runs_qm9_valid`.

Guide limitations

  • The repository warns that data preprocessing is required before training.
  • The guide documents the command workflow, not a verified training result.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
Liu-Group-UF/PropMolFlow · Official homepage title · retained in its source language
How we review sourcesSuggest a correction