Using LASSO regression to detect predictive aggregate effects in genetic studies

BMC Proceedings

Table 1 Prediction results for various model types

Model	Model type	Training A_ROC	Testing A_ROC	Number of true^a	Size^b	N
1	Genotypes only	0.57	0.55	3.57	179.43	200
	Genotypes restricted	0.56	0.55	0.84	22.07	200
	Combined model	0.82	0.82	1.27	28.38	200
	Combined model restricted	0.82	0.82	1.06	18.70	200
2a	Genotypes only	0.61	0.54	9.98	545.33	50
	Genotypes restricted	0.56	0.55	0.86	21.66	50
	Combined model	0.83	0.81	2.78	94.32	50
	Combined model restricted	0.83	0.82	1.14	20.57	50
2b	Genotypes only	0.73	0.54	11.65	348.86	150
	Genotypes restricted	0.58	0.56	2.01	29.57	150
	Combined model	0.85	0.78	9.35	228.43	150
	Combined model restricted	0.83	0.82	2.48	29.26	150
3	Genotypes only	0.62	0.54	11.32	294.68	200
	Genotypes restricted	0.58	0.56	1.75	22.84	200
	Combined model	0.83	0.82	3.94	64.17	200
	Combined model restricted	0.83	0.82	2.04	20.40	200

^a Average number of causal simulation markers included.
^b Average number of variables in each model.
Averaged results from a 5-fold evaluation procedure on N simulation data sets. Training A_ROC values were obtained from the internal 10-fold cross-validation on the training sets, as implemented in the R package glmnet. Testing A_ROC values were determined by applying each of the trained models to the five independent testing sets.

ISSN: 1753-6561