Table 4. Method comparisons for function prediction for non-synonymous mutations causing diabetes.
| Methods | Missing Rate* | False Negative Rate | False Positive Rate** | MCC |
| PhyloP | 0% | 18% | 53% | 0.300 |
| GERP++ RS | 0% | 21% | 52% | 0.281 |
| SiPhy | 0% | 16% | 51% | 0.342 |
| SIFT | 13% | 25% | 39% | 0.350 |
| PolyPhen-2 HDIV | 15% | 9% | 51% | 0.447 |
| PolyPhen-2 HVAR | 15% | 16% | 42% | 0.434 |
| LRT | 18% | 7% | 68% | 0.324 |
| MutationTaster | 3% | 3% | 77% | 0.333 |
| Mutation Assessor | 15% | 30% | 32% | 0.362 |
| FATHMM | 14% | 1% | 95% | 0.127 |
| RadialSVM score | 8% | 5% | 57% | 0.474 |
| LR score | 8% | 4% | 69% | 0.393 |
* The missing rate refers to the percentage of mutations that a method is inapplicable;
**The false positive rate was calculated by nonsynonymous single-nucleotide mutations in the diabetes genes acquired from the NHLBI GO Exome Sequencing Project (ESP) [33], the CHARGE Exome Sequencing Project [34], [36], and the 1000 Genome Project [37], excluding mutations recorded in the HGMD database.