Fig 1.
a) Known drug-target (ACT), drug-disease (ACD), and target-disease (ATD) pairwise associations are collected. b) Drug-target-disease association tensor is constructed based on ACT, ACD, and ATD. c) Multiple similarity measures for drugs, targets, and diseases are collected and are fused to build a single similarity matrix for each of drugs, targets, and diseases. d) Drug-target-disease association tensor is factorized into three factor matrices A, B, and C. e) Tensor
is reconstructed using similarity matrices upon the convergence of the factor matrices (see Section “Optimization process”). f) The pairwise or triplet association scores are computed.
Fig 2.
The effect of different values of four parameters on the performance of NTD-DR.
The panels show the AUC changes with the increase of parameter a) R b) ∝, c) λ (the same value is set to both λCT and λTD), and d) γT, respectively.
Fig 3.
ROC and precision-recall curves of NTD-DR and the existing methods for Scenario CVpairwise using dataset P.
Fig 4.
ROC and precision-recall curves of NTD-DR and the existing methods for Scenario CVcolumn-wise using dataset P.
Fig 5.
ROC and precision-recall curves of NTD-DR and the existing methods for scenario CVrow-wise using dataset P.
Table 1.
The number of known associations in the top 50 predictions made by different methods for different diseases.