Table 1.
Summary of the papers analyzed in this mini-review.
| References | Learning instances | Learning features | Machine learning algorithms | Prediction performance metrics | Results | |
|---|---|---|---|---|---|---|
| Source | Type of Feature | |||||
| Zhu et al., 2009 | DrugBank | BioGRID | Connectivity degree, cluster coefficient, distance-based measures, topological coefficient | Support Vector Machine | AUC | AUC: 69.21% |
| Jeon et al., 2014 | DrugBank, Therapeutics Target Database | Bossi and Lehner, 2009 | GARP score, RMA intensity, row chromosomal copy number, mutation occurrence and closeness centrality (combined or isolated) | SVM-recursive feature elimination (SVM-REF) method for feature selection; SVM-RBF kernels for predictions | Accuracy, Specificity, AUC | Avg. accuracy: 91.69% Avg. specificity: 91.91% Avg. AUC: 78% (combined) |
| Li et al., 2015 | DrugBank | HIPPIE | Combination of various network distance-based measures and sequence features of proteins | Random Forest with minimum Redundancy Maximum Relevance (mRMR) Feature Selection | Accuracy, Sensitivity, Specificity, Precision, Matthews correlation coefficient | Accuracy: 87.05% Sensitivity: 90.28% Specificity: 83.83% Precision: 84.82% Matthews correlation coefficient: 0.7427 (Avg. of 10 random samples) |
| Laenen et al., 2013 | PubChem, ChEMBL and BindingDB | STRING, GEO (Edgar et al., 2002) | Combination of kernel and correlation diffusion and differential gene expression | Rank-based method | AUC | Kernel: 76–91% Correlation: 89–92% |
| Emig et al., 2013 | Integrity | metaBase (Bureeva et al., 2009), GEO (Edgar et al., 2002) | Combination of neighborhood scoring, interconnectivity, network propagation, random walk and differential gene expression | Logistic regression model | AUC | AUC: 63.27–93.19% |
| Yao and Rzhetsky, 2008 | DrugBank | HPRD | Combination of connectivity, betweenness, tissue expression entropy, constant corrected ratio of non-synonymous and synonymous mutations and functional family assignment | Naive Bayesian, logistic regression, radial basis function network, Bayesian networks | AUC | Naive Bayes: 70.43% Logistic regression: 72.57% RBF network: 60.93% Bayesian Network: 72.31% |
| Costa et al., 2010 | Yildirim et al., 2007 | BioGRID, DIP, HPRD, IntAct, MINT, MIPS-MPPI, TRED, human metabolic model Recon 1 | Combination of several network measures, tissue expression profile and subcellular localization | Decision tree-based meta-classifier | AUC, Recall, Precision | AUC: 82% Recall: 78.2% Precision: 74.8% |