Topological Neural Networks go Persistent, Equivariant and Continuous

Yogesh Verma | Amauri H. Souza | Vikas Garg

The repository is developed on the intersection of RePHINE, TOGL, EMPSN and AbODE. Please refer to their repos for specific requirements.

Prerequisites

• Please install all the package requirements mentioned in RePHINE, TOGL, and AbODE.
• torchdiffeq: https://github.com/rtqichen/torchdiffeq.

Training

Graph Classification

Comparison with RePHINE

cd RePHINE/
python -u main_2d.py  --dataset {PROTEINS_full/NCI109/NCI1/IMDB-BINARY}  --gnn {gin/gcn} --diagram_type {standard/rephine}  --nsteps 20 

Comparison with TOGL

cd RePHINE/
python -u main_togl.py --dataset {ENZYMES/DD/Proteins} --gnn {gin/gcn}

QM9 Property Prediction

cd empsn/
python -u main_qm9.py --target_name {mu,alpha,gap,r2,zpve,Cv,homo,lumo} --epochs 1000 --dis 4.0 --dim 2 --num_hidden 77 --seed 42 --model_name {empsn_rephine_cont/empsn_rephine}

CDR-H3 Antibody Design

Download the train/test/val files from here. Kindly add the paths to these files in train_topnets.py file.

cd Antibody/
python -u train_topnets.py --cdr 3