Open LLM List
M P N N for molecular property prediction
keras-ioIntroduction
This repository contains the implementation of a Message Passing Neural Network (MPNN) designed to predict the blood-brain barrier permeability (BBBP) of molecules. Created by Alexander Kensert and reproduced by Vu Minh Chien, the model leverages graphs to represent molecular structures, offering an innovative approach compared to traditional methods like random forests and support vector machines.
Architecture
The MPNN model operates on undirected graphs of molecules. It uses message-passing techniques to capture intricate molecular interactions and properties, which may provide superior predictive capabilities by focusing on raw molecular features.
Training
The dataset used for training includes 2,050 molecules, each annotated with a name, label, and SMILES string. It is sourced from MoleculeNet.org. Training employs specific hyperparameters:
- Learning Rate: 0.0005
- Decay: 0.0
- Beta 1: 0.899
- Beta 2: 0.999
- Epsilon: 1e-07
- AMSGrad: False
- Training Precision: float32
Guide: Running Locally
To run the model locally, follow these steps:
- Clone the Repository: Clone the repository to your local machine.
- Install Dependencies: Use a virtual environment and install the required libraries using
pip install -r requirements.txt. - Prepare Data: Ensure your dataset is downloaded from MoleculeNet.org and properly configured.
- Execute Training: Run the training script provided in the repository.
- Evaluate the Model: Use the evaluation metrics to assess model performance.
For enhanced performance, consider using cloud GPUs such as those provided by AWS, Google Cloud, or Azure.
License
The project is shared under an open-source license, allowing for modification and distribution in accordance with the license terms. Please refer to the repository for detailed licensing information.
