Fig 1.
An illustration of the cross-network framework across two domains: drug-disease and drug-target networks. iDrug requires only partially overlapped drug nodes between these two domains. The anchor links among drug nodes are used to transfer domain knowledge across the networks.
Table 1.
Statistics of drug-disease network (Domain 1) from CTD database and drug-target network (Domain 2) from DrugBank database.
469 drugs are overlapped between two networks in total.
Fig 2.
The intra-similarity distributions in drug-disease domain.
(a) The intra-similarity distributions of drug pairs, the drug-drug similarities are calculated based on Tanimoto Score. (b) The intra-similarity distributions of disease pairs, the disease-disease similarities are computed based on the semantic similarity of MeSH terms. Note that all the self-similarity values of drugs and diseases have already been excluded in the histograms.
Fig 3.
The intra-similarity distributions in drug-target domain.
(a) The intra-similarity distributions of drug pairs, the drug-drug similarities are calculated based on Tanimoto Score. (b) The intra-similarity distributions of target pairs, the target-target similarities are calculated using the Smith-Waterman algorithm on target sequences. Note that all the self-similarity values of drugs and targets have already been excluded in the histograms. For target-target similarities, we only show the similarity values within [0, 0.2] since most of them are located in this range.
Table 2.
The symbols used in the objective function Eq (3) and their descriptions.
Fig 4.
Comparison on the performance of different methods on drug repositioning.
(a) The AUROC curves for the ‘pair prediction’ scenario. (b) The AUPR curves for the ‘pair prediction’ scenarios. (c) Precision of the top-k candidates for the ‘new drug’ scenario. (d) Precision of the top-k candidates for the ‘new disease’ scenario.
Fig 5.
Performance comparison of different methods on drug-target prediction.
(a) The AUROC curves for the ‘pair prediction’ scenario. (b) The AUPR curves for the ‘pair prediction’ scenario. (c) Precision of the top-k candidates for the ‘new drug’ scenario. (d) Precision of the top-k candidates for the ‘new target’ scenario.
Fig 6.
Performance comparison of different methods for the ‘pair prediction’ scenario on the gold standard dataset.
(a) The AUROC curves for drug-disease prediction. (b) The AUPR curves for drug-disease prediction. (c) The AUROC curves for drug-target prediction. (d) The AUPR curves for drug-target prediction.