Table 3.

Optimal feature selection and classification methods selected for each training dataset.^*

Clinical Endpoint	Train Dataset	Selection Method	Classification Method
Breast Cancer	1	Fold-Change	SVM
Estrogen	2	SAM	KNN
Receptor Status	3	mRMR	SVM
	4	Rank Prod.	LR
	5	Rank Sum	SVM
Breast Cancer	1	Rank Sum	SVM
Treat. Resp.	2	Fold-Change	Bayesian
Liver Cancer	1	Fold-Change	LR
Detection	2	SAM	SVM
	3	Fold-Change	SVM
MM Overall	1	Rank Prod.	SVM
Survival	2	Rank Sum	Bayesian
MM Event-	1	T-test	Bayesian
Free Surv.	2	SAM	Bayesian
MM Gender	1	Fold-Change	SVM
	2	T-test	LR
MM Random	1	SAM	LR
	2	Rank Prod.	SVM
NB Overall	1	Rank Prod.	SVM
Surv.	2	Rank Sum	SVM
NB Event-Free	1	mRMR	LR
Survival	2	T-test	SVM
NB Gender	1	T-test	KNN
	2	Fold-Change	SVM
NB Random	1	Fold-Change	SVM
	2	Rank Prod.	LR
Pancreatic	1	T-test	Bayesian
Cancer	2	Rank Prod.	SVM
Detection	3	SAM	SVM
	4	Fold-Change	SVM
Prostate Cancer	1	Fold-Change	Bayesian
Detection	2	Rank Sum	Bayesian
Renal Cancer	1	SAM	LR
Subtype	2	SAM	SVM
Diagnosis	3	Fold-Change	SVM
	4	Rank Prod.	LR

^*only one model is reported for each endpoint, however multiple models are possible in the event of ties