DOCSEARCHbyAlgolia
Submit Docs

Uni-Mol: Universal 3D Molecular Representation Learning

repository·main·Indexed 22 days ago

https://github.com/deepmodeling/uni-mol

A series of universal 3D molecular representation learning frameworks for drug design, quantum chemical property prediction, and protein-ligand docking. The ecosystem includes Uni-Mol for property and binding pose prediction, Uni-Mol+ for quantum chemical properties and geometry tasks, Uni-Mol2 (scalable models from 84M to 1.1B parameters), and Uni-Mol Docking V2. It also provides the `unimol-tools` Python library for automated molecular property prediction and representation.

Tokens
18.7K
Snippets
63
Records
91
Agent score
78%

What's inside Uni-Mol

  1. Overview of the Uni-Mol Framework

    main

    Uni-Mol is a universal large-scale three-dimensional Molecular Representation Learning (MRL) framework. It is designed to expand the application scope and representation capabilities of MRL through two primary models:

    1. Molecular Model: Trained on billions of molecular 3D conformations.
    2. Protein Model: Trained on millions of protein pocket data.

    Key Applications

    • Molecular Property Prediction: High performance in 3D-related tasks, including quantum chemical property prediction (via Uni-Mol+).
    • Drug Design: General applications in pharmaceutical research.
    • Material Science: Predicting properties such as gas adsorption in MOF materials and optical properties in OLED molecules.
    • Downstream Tasks: Includes docking and other molecular-related workflows.

  2. Overview of unimol_tools.data module

    main
    The unimol_tools.data module provides the core infrastructure for handling molecular data in the Uni-Mol ecosystem. It includes specialized components for loading datasets, managing data containers, scaling features, handling molecular conformers, and managing dataset splits (e.g., train/test/validation).

  3. Overview of the Uni-Mol series methods

    main

    The Uni-Mol repository provides several specialized frameworks for molecular modeling:

    • Uni-Mol: A universal 3D molecular representation learning framework for tasks like property prediction and binding pose prediction.
    • Uni-Mol+: Designed for quantum chemical property prediction and molecule geometry tasks (conformation generation/optimization).
    • Uni-Mol Tools: High-level Python wrappers for automated property prediction and representation using Uni-Mol.
    • Uni-Mol Docking V2: Tools for realistic and accurate protein-ligand binding pose prediction.
    • Uni-Mol2: A scalable molecular pretraining model with sizes ranging from 84M to 1.1B parameters.

  5. Use Uni-Mol tools for molecular tasks

    main

    The Uni-Mol tools package provides easy-to-use wrappers for integrating Uni-Mol into various workflows. You can use these tools for:

    • Molecular property prediction: Predicting specific characteristics of molecules using Uni-Mol.
    • Molecular representation: Extracting 3D molecular representations.
    • Downstream tasks: Utilizing Uni-Mol for more complex molecular modeling workflows.

  12. Install Uni-Mol from source

    main

    If you prefer to install from the source code, follow these steps:

    1. Install dependencies from requirements.txt.
    2. Clone the repository.
    3. Navigate to the unimol_tools directory.
    4. Run the setup script.
    pip install -r requirements.txt
git clone https://github.com/deepmodeling/Uni-Mol.git
cd Uni-Mol/unimol_tools
python setup.py install