Label Propagation Prediction of Drug-Drug Interactions Based on Clinical Side-Effects




·      Known DDIs

We used the unsafe co-prescriptions from TWOSIDES as known set of DDIs. There are 645 drugs and 63,473 distinct pairwise DDIs in the dataset.

TWOSIDES was obtained from Tatonetti Lab.


·      Label side effects

We used the entire SIDER database which contains package insert information on market medicines and their recorded adverse drug reactions. There are 996 drugs and 4,192 side effects in the dataset.

SIDER was obtained from SIDER2 Side Effect Resource.


·      Off-label side effects

We used OFFSIDES database which built by mining FAERS system while controlling clinical confounding factors. There are 1,332 drugs and 10,093 side effects in the dataset.

OFFSIDES was obtained from Tatonetti Lab.


·      Chemical structures

All chemical structures were obtained from PubChem.

STITCH id used by SIDER, OFFSIDES, and TWOSIDES was mapped to PubChem compound id based on this rule.

The descriptions of the 881 PubChem chemical substructures found here.


Processed Data for Evaluation Experiments:

·      Name of 569 drugs in our validation experiments: Download

·      DDIs among the 569 drugs: Download

·      Label side effect similarity scores: Download

·      Off-label side effect similarity scores: Download

·      Chemical structure similarity scores: Download


Novel Predictions:

·      Name of 1626 drugs with one or more side effect profiles: Download

·      Predicted drug-drug interactions: Download

It includes 145,068 predicted DDIs by our method.