Table 1.
Classic ML methods to predict protein–ligand binding sites
| SN | Approach | Techniques | Features | Database used | Year |
|---|---|---|---|---|---|
| 1 | Oriented Shell Model [73] | Support vector machine | Developed oriented shell model, utilizing distance and angular position distribution | Self-curated | 2005 |
| 2 | SitePredict [74] | Random forest | Predicted small ligand-binding sites mobilizing backbone structure | Self-curated | 2008 |
| 3 | LIBRUS [75] | Support vector machine | Combined ML and homology information for sequence-based ligand-binding residue prediction | Self-curated + FINDSITE’s database | 2009 |
| 4 | Qiu and Wang’s method [76] | Random forest | Used eight structural properties to train random forest classifiers, latter combined to predict binding residues | Q-SiteFinder’s dataset | 2011 |
| 5 | Wong et al.’s method [77] | Support vector machine + differential evolution | Classified the grid points with the location most likely to contain bound ligands | LigASite | 2012 |
| 6 | DoGSiteScorer [78] | Support vector machine | Web server for binding site prediction, analysis and druggability assessment | Self-curated | 2012 |
| 7 | Wong et al.’s method [79] | Support vector machine | Used SVM to cluster most probable ligand-binding pockets using protein properties | LigASite + self-curated | 2013 |
| 8 | TargetS [80] | Support vector machine + modified AdaBoost | Designed template-free predictor with classifier ensemble and spatial clustering | BioLip | 2013 |
| 9 | Wang et al.’s method [81] | Support vector machine + statistical depth function | SVM model integrating sequence and structural information | PDBbind | 2013 |
| 10 | LigandRFs [82] | Random forest | Applied random forest ensemble to identify ligand-binding residues from sequence information alone | CASP9 targets + CASP8 targets | 2014 |
| 11 | Suresh et al.’s method [83] | Naive Bayes classifier | Trained Naive Bayes classifier using only sequence-based information | Self-curated | 2015 |
| 12 | OSML [84] | Support vector machine | Proposed dynamic learning framework for constructing query-driven prediction models | BioLip + CASP9 targets | 2015 |
| 13 | PRANK [7] | Random forests | Developed mechanism to prioritize the predicted putative pockets | Astex Diverse set + self-curated | 2015 |
| 14 | UTProt Galaxy [85] | Support vector machine + neural network + random forest | Developed pipeline for protein–ligand binding site predictive tools using multiomics big data | Self-curated | 2015 |
| 15 | Chen et al.’s method [86] | Random forest | Proposed dynamic ensemble approach to identify protein–ligand binding residues by using sequence information | ccPDB + CASP9 targets + CASP8 targets | 2016 |
| 16 | Chen et al.’s method [87] | Random forest | Predicted allosteric and functional sites on proteins | PDBbind + allosteric DB + CATH DB | 2016 |
| 17 | TargetCom [88] | Support vector machine + modified AdaBoost algorithm | Designed ligand-specific methods to predict the binding sites of protein–ligand interactions by an ensemble classifier | BioLip | 2016 |
| 18 | P2Rank 2.1 [89] | Bayesian optimization | Improved version of P2Rank | Self-curated | 2017 |
| 19 | P2Rank [90] | Random forest | Built stand-alone template-free tool for prediction of ligand-binding sites | Self-curated | 2018 |
| 20 | PrankWeb [91] | Random forest | Online resource providing an interface to P2Rank | Self-curated | 2019 |