Molecular-toxicity-prediction

1. Knowledge Graph Deployment

All ToxKG data are stored in a neo4j.dump file. To deploy it locally, follow the steps below (note: importing a dump file requires using the same major version of Neo4j as our build, i.e., Neo4j Server 5.x).

neo4j stop
neo4j-admin database load neo4j --from=/path/neo4j.dump --force
neo4j start

2. Environment Setup

The environment used in this project has been fully encapsulated in the environment.yml file. You can install it with a single command as follows:

conda env create -f environment.yml -n myenv
conda activate myenv

3. Code Usage Instructions

3.1 File Description

data directory

Location for storing data and input features.

data/KG/r-gcn directory

Contains data files related to the R-GCN model.

common_cids.csv     --Information containing the CIDs, SMILES representations, and the 12 toxicity prediction results of the common molecules shared by Tox21 and the knowledge graph
gnn_input_fp.csv    --Information containing CIDs and five different molecular fingerprints
filtered_triples.csv    --Complete triplet information of molecules with shared CIDs
gnn_input_genes.csv     --Gene list file
gnn_input_pathways.csv  --File containing the pathway list
compound_master.csv --Contains CID, SMILES, {fingerprint_cols…}, and {12 toxicity columns}

GAT/GCN/GPS/HGT/HRAN/R-GCN

Directory containing the training code and training results for the six GNN models.

save folder

Stores the training results.

train folder

Contains the training code.

data_utils

Contains data preprocessing scripts.

4. Model Training Commands

4.1 GCN Execution Command

python gcn2-bac.py --epochs 1000 --hidden 512 --heads 4

4.2 GAT Execution Command

python gat2-bac.py --epochs 1000 --hidden 512 --heads 4

4.3 GPS Execution Command

python train_gps_gine_cv.py --epochs 1000 --hidden 512 --heads 4

4.4 HGT Execution Command

python train_hgt_cv.py --epochs 1000 --hidden 128 --lr 5e-4 --embed_dim 32

4.5 HRAN Execution Command

python train_hran_pos_weight_cv.py \
       --gene_col geneSymbol --pathway_col pathwayId \
       --epochs 1000 --hidden 512

4.6 R-GCN Execution Command

python train_rgcn_cv2.py \
       --gene_col geneSymbol --pathway_col pathwayId \
       --epochs 1000 --hidden 512