For any query drug d, its side effects were determined by the following strategy: (1) if there exist interactive compounds of d in the training set, only chemical-chemical interactions were used to identify its side effects; (2) otherwise, both chemical-chemical interactions and protein-chemical interactions were employed to make prediction.
According to (10), the greater score ∏ c (d → C j ) means that there are lots of interactive compounds of d that have the side effect C j or some interactions between d and its interactive compounds with the side effect C j are labeled by high confidence scores.
Under such circumstance, there were no interactive compounds of d q in the training set and no meaningful result can be obtained by this method.
Since side effect is one part of biological functions of drugs, using the properties of interactive compounds to identify drugs side effects is a feasible scheme.
In view of this, it may be feasible to utilize the information of interactive compounds to predict diseases that a query drug can treat.
To obtain the information of interactive compounds, we downloaded the data of chemical-chemical interactions from STITCH (http://stitch.embl.de/, chemical_chemical.links.detailed.v3.0.tsv.gz) [ 12], a well-known database containing known and predicted chemical-chemical interaction and protein-chemical interaction data from experiments, literature, or other reliable sources.
The information of interactive compounds was downloaded from STITCH (chemical_chemical.links.detailed.v3.1.tsv.gz, http://stitch.embl.de/) [ 18], a well-known database containing the interaction information of chemicals and proteins.
Both the first- and the second-step mutants exhibited increased resistance to a number of antimitotic drugs (viz. vinblastine, vincristine, colchicine, taxol and maytansine), several protein synthesis inhibitors (viz. chalcomycin, bruceantin, harringtonine, homoharringtonine), a large number of DNA interactive compounds (viz.
The interactive compounds thus considered here satisfy one of the following three properties: (I) they participate in the same reactions; (II) they share similar structures or activities; (III) they have literature associations.
Since the last score combines the information of other scores, we utilized the last score to indicate the interactivity of two chemicals in this study; that is, compounds in the interaction unit with "Combined_score" greater than zero were deemed to be interactive compounds.
In view of this, the target group of a query drug d q can be determined by its interactive compounds in the training set.
