CB-TIP is a graph-based multi-relational link prediction framework tailored for DDI prediction, which is useful for providing prescript tips to clinicians. See our paper for details on the framework.

You need to create a CONDA environment with the command conda env create -f torch17_chembiotip.yaml.

We provide two preprocessed real-world datasets C-DB and CB-DB (path: data) used in our paper. To reproduce the results reported in our paper, there is no need to construct the intra-view graphs from the raw linear expressions of molecules. We share the archives of LMDB files storing the built intra-view graphs (path: data/DATASET_NAME/lmdb_files).

*If you want to run this code on your own dataset,

1. Generate the residue contact maps for large molecules (please refer to Section 7.1 and the files i)utils/msa_aln_gen.py ii) utils/cmap_gen.py).
2. Organize your data as the form like data/CB-DB/.
3. Construct intra-view graphs and save them into LMDB files with the code utils/generate_intra_graph_db.py.

1. Run data/calc_fp.py to generate fingerprints for chemical drugs and divide them into clusters.
2. Run data/construct_pp_nn.py to similar pairs and dissimilar pairs of chemical drugs.

To run the training process under default settings, use the following command:

$ python train_main.py -d DATASET_NAME -sp SPLIT

options for DATASET_NAME: C-DB, CB-DB Negative samples: You can use the negative samples in data/DATASET_NAME/ddi_neg.csv, or generate negative samples using the code utils/sample_neg_split.py. SPLIT: You can split the positive/ negative samples into training/ validation/ test sets using the code utils/sample_neg_split.py. Please remember to update the data path in the function dd_dt_tt_build_inter_graph_from_links in utils/hete_data_utils.py.

For more customized settings, please refer to utils/arg_parser.py.

If you use CB-TIP in your research, please use the following BibTeX entry. 📣 Thank you!

@article{10.1093/bib/bbad271,
    author = {Ru, Zhongying and Wu, Yangyang and Shao, Jinning and Yin, Jianwei and Qian, Linghui and Miao, Xiaoye},
    title = "{A dual-modal graph learning framework for identifying interaction events among chemical and biotech drugs}",
    journal = {Briefings in Bioinformatics},
    pages = {bbad271},
    year = {2023},
    month = {07},
    issn = {1477-4054},
    doi = {10.1093/bib/bbad271}
}