Uni Mol Models

Introduction

The Hugging Face repository for Uni-Mol models offers a suite of pretrained models utilizing Uni-Mol, a universal framework for 3D molecular representation learning. These models are designed for various applications, including molecular property prediction and materials science.

Architecture

Uni-Mol models are pretrained with specific molecular data, catering to different scientific needs:

• mol_pre_all_h_220816.pt: Includes hydrogen atoms for detailed molecular structure analysis.
• mol_pre_no_h_220816.pt: Excludes hydrogen atoms, focusing on core molecular structures.
• mp_all_h_230313.pt: Trained on crystalline material data for materials science applications.
• oled_pre_no_h_230101.pt: Tailored for OLED technologies, excluding hydrogen atoms.
• poc_pre_220816.pt: Focused on protein pocket interactions, aiding drug discovery.

Each model has a corresponding dictionary file (*_dict.txt) for accurate data mapping.

Training

The models are pretrained on diverse datasets, with some incorporating hydrogen atoms while others do not, depending on the intended application. The training data and techniques are selected based on the specific molecular or material properties the model is designed to predict.

Guide: Running Locally

1. Clone the GitHub Repository: Visit the Uni-Mol GitHub Repository to download the models and access detailed instructions.
2. Install Dependencies: Follow the setup instructions provided in the repository to install necessary dependencies.
3. Load Pretrained Models: Use the provided examples to load models into your application.
4. Run Predictions: Utilize the models for predictions based on your data requirements.

For optimal performance, it is recommended to use cloud GPUs from providers like AWS, Google Cloud, or Azure, especially for large-scale predictions or training.

License

This project is licensed under the MIT License, allowing for flexible use and distribution.