Table 1.
Performance assessment of in silico prediction tools on experimentally validated variants (n = 257).
| Scoring metric | n Missing | Sensitivity | Specificity | Accuracy | PPV | NPV |
|---|---|---|---|---|---|---|
| HSF (2%) | 28 | 0.8941 | 0.3958 | 0.5808 | 0.4663 | 0.8636 |
| SpliceAI (0.2) | 11 | 0.8987 | 0.9162 | 0.9106 | 0.8353 | 0.9503 |
| Alamut SSF (5%) | 5 | 0.7317 | 0.9294 | 0.8651 | 0.8333 | 0.8778 |
| Alamut MES (10%) | 1 | 0.7381 | 0.9070 | 0.8516 | 0.7949 | 0.8764 |
| Alamut NNSplice (5%) | 11 | 0.6923 | 0.8631 | 0.8089 | 0.7013 | 0.8580 |
| Alamut 2/3 | 14a | 0.7237 | 0.9162 | 0.8560 | 0.7971 | 0.8793 |
Values have been calculated omitting the missing scores for each tool.
HSF Human Splicing Finder, MES MaxEntScan, NPV negative predictive value, PPV positive predictive value, SSF Splice Site Finder.
a11 variants missing one score, three variants missing two scores.