SpecGOFiller: Species-Aware Multi-Network Learning Improves Function Prediction in Partially Annotated Proteins
SpecGOFiller, a species-aware multi-network graph learning framework. SpecGOFiller jointly models sequence, interaction, and functional similarity networks within a unified graph convolutional architecture, while explicitly encoding species identity to capture functional dependencies across the tree of life.
Our model is implemented by Python 3.8 with Pytorch 2.3.1 and Pytorch-geometric 2.5.3, and run on Nvidia GPU with CUDA 12.4.
Please download the data (https://drive.google.com/drive/folders/1pK8IY9_DaVqgkyLibfv2D8LTqiZD_4hK?usp=share_link) and put them into the data directory,
Run the following command to train the model:
python main.py