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

Shepherd

ShEPhERD is an open-source Python diffusion model with code for training on molecular structures and sampling new 3D conformations conditioned on shape, electrostatics and pharmacophore interaction profiles.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Shepherd — 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 ShEPhERD when the task is research-oriented 3D molecular generation conditioned on target interaction features, rather than general-purpose text or image generation.

Who it is for

  • Researchers exploring computational drug design and bioisosteric molecular candidates.
  • Developers building or adapting research workflows for 3D molecular generation.
  • Teams training diffusion models on molecular conformer data.

Documented uses

  • Generate 3D molecular candidates that exhibit a specified target interaction profile.
  • Train a diffusion model over molecular structures and selected interaction features.
  • Load pretrained models for generation with different conditioning combinations.
Before you choose

Requirements & limits

The repository is public and MIT-licensed. It is self-hosted Python software; pretrained weights are downloaded from Hugging Face and cached locally.

  • The available pretrained model types support different conditioning inputs, so select one that matches the target profile needed.
  • The README lists Python 3.9 to below 3.12, RDKit, PyTorch, Open3D and xTB requirements; PyTorch Geometric installation is also documented.
  • The repository notes that its stored checkpoints required about two weeks of training on two V100 GPUs, indicating substantial compute for training.
  • shepherd-score is a distinct sister repository for conformer operations, profile extraction, alignment, scoring and evaluation, with separate installation requirements.

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

Load a pretrained ShEPhERD model

Load the default MOSES-aq checkpoint or select a documented model type in Python.

Before you start

  • ShEPhERD installed in a Python environment
  • A model type supported by the repository, such as the default mosesaq or gdb_x3
  • Network access if the required model weights are not already cached

Steps

  1. Import load_model from shepherd.

  2. Call load_model() for the default MOSES-aq model, or pass a model type such as gdb_x3.

  3. Allow the model weights to download from Hugging Face if needed; they are cached locally.

Expected outcome

A model object is loaded for the selected model type, with weights downloaded and cached when necessary.

Guide limitations

  • The documented model types differ in conditioning: mosesaq uses shape, electrostatics, and pharmacophores; gdb_x3 uses shape and electrostatics.
  • The loading example does not document a generation call or describe how to construct inference inputs.
  • Older repository versions may require manually obtaining checkpoints; the README recommends its loading functions for ShEPhERD versions greater than 0.2.4.
Documentation 1 ↗
Evidence and transparency

Sources & review

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