Skip to content

GraphAF

GraphAF is a flow-based autoregressive model for molecular graph generation, with a reference-code link and a README update pointing to an implementation in TorchDrug.

Catalog updated ·

Website ↗GitHub / Repository ↗Documentation ↗
Follow

Public link：https://chemaiatlas.com/models/graphaf/

Claim this resource

Overview

GraphAF is a research model for molecular graph generation. The supplied project README identifies it as a flow-based autoregressive model and connects the reference implementation to the paper “GraphAF: a Flow-based Autoregressive Model for Molecular Graph Generation,” presented at ICLR 2020. Its directory role is therefore a molecular-generation model rather than a general-purpose chemistry toolkit or an independently documented service.

The README provides two routes for investigating implementation details: a Google Drive link described as containing the code, and an update linking to a GraphAF implementation in TorchDrug. These are distinct access paths. The repository page serves as a short project pointer, while the TorchDrug link leads to a generation tutorial in that platform. The supplied excerpt does not establish whether the two implementations have identical interfaces, dependencies or behavior.

The documented output domain is molecular graphs. However, the excerpt does not specify training inputs, graph encodings, conditioning options, output serialization or a runnable generation procedure. It also supplies no performance measurements, installation instructions, checkpoint details or code-licence text. Researchers considering GraphAF for a generation workflow should use the linked paper and implementation resources to establish those details before designing an evaluation. The source supports the model’s identity and implementation pointers, but not claims about generation quality, chemical validity or suitability for a particular discovery task.

Key Features

  • Uses a flow-based autoregressive modeling approach for molecular graph generation.
  • Is associated with a named ICLR 2020 research paper, linked directly from the project README.
  • Provides a Google Drive access link for the reference code.
  • Points to an implementation of GraphAF in TorchDrug through a linked generation tutorial.

Use Cases

  • Intended evaluation: examine GraphAF as a candidate model for a molecular graph generation research workflow, after confirming the implementation’s input and output formats.
  • Intended evaluation: compare the reference-code route with the TorchDrug implementation route to determine which fits an existing research environment.
  • Research orientation: use the linked paper and implementation pointers to scope a reproduction study without treating the short README as a complete experimental protocol.

How to Use

  1. Start with the project README to confirm the model identity and distinguish its reference-code link from the later TorchDrug implementation pointer.
  2. Consult the GraphAF paper for the methodological context. Identify the assumptions and experimental setup relevant to your intended evaluation; these details are not reproduced in the supplied excerpt.
  3. Follow the reference-code link and inspect any available instructions, dependencies and licence information before attempting to run it.
  4. Inspect the TorchDrug generation tutorial as an alternative implementation route. Confirm its actual interface and requirements rather than assuming equivalence with the reference code.
  5. Before evaluating either route, document the supported molecular graph representation, required inputs, generated outputs and evaluation criteria. Treat execution and scientific validation as work still to be performed.

Related resources

AIDDISON Explorer

Platforms

AIDDISON Explorer is a hosted drug-discovery platform that generates and ranks molecular candidates against target profiles, design constraints, predicted properties and synthetic feasibility.

Molecular Generation · Drug Discovery

Updated

Chemistry42

Platforms

Chemistry42 is a small-molecule discovery platform combining generative design, retrosynthesis, ADMET and selectivity prediction, and physics-based prioritization for hit identification and lead optimization.

Molecular Generation · Drug Discovery

Updated

Equivariant Diffusion Models

Model

Official research code for E(3)-equivariant diffusion models that generate 3D molecules, with QM9 and GEOM-Drugs training workflows, sample analysis, and property-conditioned generation.

Open sourcePython

Molecular Generation

Updated

FEgrow

Open Source

FEgrow supports interactive ligand-series construction for free-energy preparation, with documented active-learning examples and Dask-based acceleration for molecular design workflows.

Open source

Molecular Generation · Drug Discovery

Updated

GeoDiff

Model

GeoDiff is a geometric diffusion model for molecular conformation generation, with official code for GEOM-based training, checkpoint sampling, and conformation and property evaluation.

Open sourcePython

Molecular Generation · Computational Chemistry

Updated

GEOM

Dataset

GEOM provides 37 million energy- and statistical-weight-annotated molecular conformations for over 450,000 molecules, with MessagePack data, RDKit objects, and loading and analysis tutorials.

Python

Molecular Generation · Computational Chemistry

Updated

Related guides

Workflows

AI for Drug Discovery: Agents, MCP Servers and Platforms

Choose tools by workflow stage: target evidence, protein structures, known binding sites, molecular generation, property analysis and screening. Compare research agents, MCP integrations, hosted APIs and platforms, with a proposed end-to-end example and practical evaluation gates.

· 10 min read
Workflows

Reproducibility for Chemistry AI Workflows

Build a compact evaluation record that connects scientific questions to data provenance, transformations, model configurations, outputs and failures, with practical planning examples for chemistry AI.

· 6 min read