Methodology for biomarker discovery with reproducibility in microbiome data using machine learning

BMC Bioinformatics

Table 2 Individual and average AUCs and MCCs from the validation phase and the additional validation approaches applied to the IBD datasets. The standard deviation of each result was excluded to keep the table simple and avoid complexity

PRJEB21504*	53 features (REFS)		1793 features		SelectKbest (k = 53)
Classifier	AUC	MCC	AUC	MCC	AUC	MCC
AdaBoostClassifier	0.9100	0.8623	0.9100	0.8337	0.9100	0.8577
Extra Trees	0.9000	0.8841	0.8600	0.8049	0.9400	0.8577
KNeighbors	0.9300	0.8547	0.5400	0.0845	0.8900	0.8353
MLP	0.9900	0.9900	0.6100	0.2640	0.8900	0.8767
Lasso CV	0.9500	0.7564	0.6700	0.4064	0.8800	0.7165
Average	0.9360	0.8715	0.7180	0.4787	0.9020	0.8287

DRA006094	22 of 53 features (REFS)		SelectKbest (21 of 53 features)		10-time random selection
Classifier	AUC	MCC	AUC	MCC	AUC	MCC
AdaBoostClassifier	0.7100	0.3087	0.5190	0.3288	0.7800	0.0761
Extra Trees	0.7800	0.4585	0.5210	0.3881	0.7200	0.0599
KNeighbors	0.8300	0.4245	0.5260	0.4070	0.6800	0.0093
MLP	0.8300	0.4418	0.5510	0.3881	0.7200	0.0916
Lasso CV	0.7400	0.3952	0.5230	0.4151	0.7600	0.0433
Average	0.7780	0.4057	0.5280	0.3854	0.7320	0.0560

PRJNA684584	48 of 53 features (REFS)		SelectKbest (52 of 53 features)		10-time random selection
Classifier	AUC	MCC	AUC	MCC	AUC	MCC
AdaBoostClassifier	0.7200	0.4151	0.5410	0.3364	0.6400	0.1112
Extra Trees	0.7500	0.4300	0.5600	0.4026	0.7400	0.1612
KNeighbors	0.7000	0.3111	0.5610	0.1081	0.5900	0.1392
MLP	0.6800	0.3657	0.5700	0.2694	0.6800	0.1393
Lasso CV	0.7000	0.2616	0.5590	0.2908	0.6100	0.1420
Average	0.7100	0.3567	0.5582	0.2814	0.6520	0.1386

ISSN: 1471-2105