Large sample size and nonlinear sparse models outline epistatic effects in inflammatory bowel disease

Table 2 Ensemble model

Model	ROC AUC^a	Parameters
NN_biosparse	0.758 (0.00689)	25,503
Randomly sparsified (100% genes connected)	0.759 (0.00778)	25,503
Randomly sparsified (75% genes connected)	0.760 (0.00571)	19,708
Ensemble (100% genes connected)	0.774 (0.00276)	25,503
Ensemble (75% genes connected)	0.776 (0.00265)	19,709

^aPerformance given as mean (standard deviation) of test set ROC AUC from 10 different full threefold cross-validation runs

ISSN: 1474-760X