Skip to main content

Table 4 Standard performance benchmarks for MutPred Splice based on an unseen test set of 352 variants (238 positive, 114 negative) using the three different iterations (Iter. 1, Iter 2. and Iter. 3) of the four different training sets identified in this study (Table  2 )

From: MutPred Splice: machine learning-based prediction of exonic variants that disrupt splicing

Data set False positive rate (%) Sensitivity (%) Specificity (%) Accuracy (%) AUC (%) MCC
Disease negative set Iter. 1 7.0 53.4 93.0 73.2 75.2 0.45
Iter. 2 7.0 52.5 93.0 72.8 75.9 0.44
Iter. 3 4.4 55.0 95.6 75.3 77.1 0.49
SNP negative set Iter. 1 36.8 73.1 63.2 68.1 76.4 0.35
Iter. 2 36.8 72.3 63.2 67.7 76.8 0.34
Iter. 3 34.2 71.0 65.8 68.4 78.3 0.35
Mixed negative set Iter. 1 7.9 56.3 92.1 74.2 78.8 0.46
Iter. 2 7.9 56.7 92.1 74.4 78.6 0.46
Iter. 3 7.0 64.7 93.0 78.8 83.5 0.54
Random SNP set Iter. 1 0.0 1.3 100.0 50.6 50.6 0.06
Iter. 2 0.9 1.7 99.1 50.4 45.2 0.03
Iter. 3 29.8 31.1 70.2 50.6 50.3 0.01
  1. Classification models were built using RF with 1,000 trees. The unseen test set was experimentally characterized with respect to the splicing phenotype. Performance benchmarks for the final classification model (Mixed negative set; Iter. 3) are highlighted in bold. Performance metrics where appropriate were calculated using a probability threshold (general score) ≥0.60. The Random SNP set is a control set. MCC, Matthews correlation coefficient.