MolGraph-xLSTM is a graph-based dual-level xLSTM framework designed for molecular property prediction. This model improves molecular feature extraction and enhances prediction accuracy for both classification and regression tasks.

To ensure reproducibility, all dependencies are listed in requirements.txt. Below are the tested installation steps for setting up the environment on Linux (Ubuntu 22.04) using Conda and Python 3.10.0.

Clone the repository and set up the Conda environment:

git clone https://github.com/syan1992/MolGraph-xLSTM
cd MolGraph-xLSTM

conda create -n molgraph-xlstm python=3.10.0 -y
conda activate molgraph-xlstm

pip install --index-url https://download.pytorch.org/whl/cu121 torch==2.2.0+cu121 torchvision torchaudio
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv -f https://data.pyg.org/whl/torch-2.2.0+cu121.html
pip install -r requirements.txt

The following datasets are used in our experiments:

1. MoleculeNet Dataset

Each dataset was split into training (80%), validation (10%), and test (10%) subsets.
All partitioned datasets are located in the datasets folder.

2. TDC Dataset

Each dataset was split into training (70%), validation (10%), and test (20%) subsets.
All partitioned datasets are located in the datasets folder.

1. Classification
   python main.py --lr_decay_epochs 300 --lr_decay_rate 0.5 --learning_rate 0.0001 --batch_size 128 --epochs 100 --trial 44 --dataset bace --num_tasks 1 --classification --num_blocks 2 --slstm 0 --data_dir "datasets" --num_gc_layers 4 --power 4 --num_dim 256 --mlp_layer 2 --num_experts 8 --num_heads 16
2. Regression
   python main.py --lr_decay_epochs 800 --lr_decay_rate 0.5 --learning_rate 0.0002 --batch_size 128 --epochs 200 --trial 41 --dataset freesolv --num_tasks 1 --num_blocks 2 --slstm 0 --data_dir "datasets" --num_gc_layers 4 --power 4 --num_dim 256 --dropout 0.5 --mlp_layer 1 --num_experts 8 --num_heads 8

Run the commands in Running Code according to the following hyperparameter settings for other datasets:

Moleculenet Dataset Hyperparameter Setting

TDC Dataset Hyperparameter Setting

This project is licensed under the MIT License.

Supervised contrastive learning : https://github.com/HobbitLong/SupContrast
mixture-of-expert: https://github.com/lucidrains/mixture-of-experts/tree/master