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

ShEPhERD-2

ShEPhERD-2 is an open-source Python model and codebase for generating 3D molecular structures conditioned on interaction profiles combining shape, electrostatics and pharmacophores. It includes controls for pharmacophore prioritization, substructure constraints and composition of multiple profiles, supporting molecular-design explorations such as fragment merging and dual-target design.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

ShEPhERD-2 — 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-2 when a molecular-generation workflow can be guided by 3D interaction features and you need controls such as scaffold constraints, pharmacophore prioritization or combined profiles.

Who it is for

  • Researchers exploring generative molecular design from 3D structures or molecular interaction profiles.
  • Python users who want command-line or programmatic access to conditioned molecule generation.
  • Model developers interested in training or evaluating the model using the repository's scripts and data formats.

Documented uses

  • Generate diverse molecular candidates conditioned on a reference structure's shape, electrostatics and pharmacophores.
  • Constrain generation around selected scaffold atoms or prioritized pharmacophores.
  • Explore molecule-design cases involving bioisosteric fragment merging, dual-target design, selectivity engineering or modality hopping.
  • Run model training with the supplied training script and configured HDF5 data.
Before you choose

Requirements & limits

The code is public under the MIT license. The documented setup uses Python 3.11 and xTB; trained model weights are downloaded automatically on first use.

  • The repository's suggested environment uses Python 3.11 and a CUDA-specific package extra for GPU installation.
  • xTB is a required external dependency and is used for the documented relaxation of SMILES-derived inputs and generated samples.
  • SDF input is recommended and preserves supplied poses by default; SMILES inputs are embedded and relaxed.
  • The repository documents ARM-specific wheel considerations when selecting CUDA versions above 12.4.

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

Generate candidates from a molecular reference

Use the documented command-line inference workflow to generate molecular candidates conditioned on a reference structure or SMILES.

Before you start

  • A Linux environment with Python 3.11, as used in the repository's suggested setup.
  • The ShEPhERD-2 repository and its Python dependencies installed in a conda environment.
  • xTB installed and available on PATH; the repository states it is required and is not pip-installable.
  • One input: an SDF structure, SMILES string, SMILES file, or molblock-and-partial-charge pickle.

Steps

  1. Clone the repository and enter its directory.

  2. Create and activate a Python 3.11 conda environment with xTB, then install the project using the documented CUDA-specific package extra.

  3. Run the command below with one supported input option and --output-dir; the README gives SDF and SMILES examples.

    Documented command
    python generate.py

  4. Review the generated SDF files under the output directory's results/reference_NNN/molecules.sdf path.

Expected outcome

Valid generated molecules are saved as SDF files in the specified output directory's results subfolder.

Guide limitations

  • The documented installation example uses a CUDA-specific extra; the README gives separate wheel guidance for ARM systems.
  • SMILES inputs are embedded and xTB-relaxed. SDF poses are preserved by default and can optionally be relaxed.
  • The repository says trained weights download automatically on first use.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
15 October 2026
Original website title
coleygroup/shepherd2 · Official homepage title · retained in its source language
How we review sourcesSuggest a correction