Table 1.
Features, Experimental Evaluation Results and Assessment of Commonly Used Algorithms in miRNA Target Prediction
| Target prediction algorithm | Features | Experimental evaluation results | Assessment | Reference | |||||
|---|---|---|---|---|---|---|---|---|---|
| Parameters contributing to the final score | Cross-species conservation | Sethupathy et al. 2006 | Baek et al. 2008 | Alexiou et al. 2009 | Advantages | Disadvantages | |||
| sensitivity1 | log2-fold change2 | precision3 | sensitivity4 | ||||||
| miRanda | complementarity and free energy binding | conservation filter is used | 49% | 0.14 | 29% | 20% | - beneficial for prediction sites with imperfect binding within seed region | - low precision, too many false positives | [56] |
| TargetScan | seed match, 3’ complementarity local AU content and position contribution5 | given scoring for each result | 21% | 0.326 | 51% | 12% | - many parameters included in target scoring - final score correlates with protein downregulation |
- sites with poor seed pairing are omitted | [38] |
| TargetScanS | seed match type | only conservative sites are considered | 48% | – | 49% | 8% | - simple tool for search of conserved sites with stringent seed pairing | - underestimate miRNAs with multiple target sites | [38] |
| PicTar | binding energy, complementarity and conservation | required pairing at conserved positions | 48% | 0.26 | 49% | 10% | - miRNAs with multiple alignments are favored | - does not predict non-conservative sites | [42, 57] |
| DIANA-microT | free energy binding and complementarity | dataset of conserved UTRs among human and mouse is used | 10% | – | 48% | 12% | - SNR ratio and probability given for each target site - possibility of using own miRNA sequence as an input |
- some miRNAs with multiple target sites may be omitted | [36] |
| PITA | target site accessibility energy | user-defined cut-off level |
– | 0.046 | 26% | 6% | - the secondary structure of 3’UTR is considered for miRNA interaction | - low efficiency compared to other algorithms | [46] |
| Rna22 | pattern recognition and folding energy | not included | – | 0.09 | 24% | 6% | - allows to identify sites targeted by yet-undiscovered miRNAs | - low efficiency compared to other algorithms | [58] |
percentage of experimentally supported miRNA-target gene interactions predicted (used TarBase records for which a direct miRNA effect was examined).
average protein depression of genes predicted by the algorithm to be miR-223 targets.
proportion of correctly predicted target miRNAs to total predicted miRNA-mRNA interactions (data obtained from proteomic analyses carried out by Selbach et al.).
proportion of correctly predicted target miRNAs to total correct miRNA-mRNA interactions (data obtained from proteomic analyses carried out by Selbach et al.).
position contribution parameter promotes sites close to the 3’UTR ends.
the final scoring correlates with the level of protein downregulation.