Problems with the nested granularity of feature domains in bioinformatics: the eXtasy case

BMC Bioinformatics

Table 1 Results of the eXtasy-based benchmark.

	eXtasy benchmark class distribution		realistic class distribution
Metric	bootstraping	hierarchical sampling	bootstraping	hierarchical sampling
Sensitivity	0.863751*	0.783985	0.888667*	0.816667
Specificity	0.951801	0.980001*	0.951809	0.980030*
Precision	0.711571	0.842510*	0.002423	0.005313*
NPV	0.980934*	0.970764	0.999985*	0.999976
MCC	0.751416	0.788129*	0.044896	0.064834*
AUC-ROC	0.972466	0.973497	0.979336	0.981345
AUC-PR	0.888809	0.895255	0.095083	0.128191

Results of the benchmark in terms of average Sensitivity, Specificity, Precision, Negative predictive value (NPV) and Matthews correlation coefficient (MCC). Bold typing indicates which method of generating in-bag sample is better given a performance measure and a classification scenario, while asterisk (*) points on statistically significant differences (p = 0.05 level). The values of the area under the ROC and PR curve are obtained from the average curves constructed by threshold averaging; hence no statistical testing has been applied on these.

ISSN: 1471-2105