LightCUD: a program for diagnosing IBD based on human gut microbiome data

Table 1 Comparison of model performances built with five different machine learning algorithms. LightGBM performed better than the other four algorithms, with the highest AUC in all the four discrimination tasks (health vs IBD and UC vs CD) and the highest AP in three out of four tasks

Discrimination tasks	WGS Health vs IBD		WGS UC vs CD		16S Health vs IBD		16S UC vs CD
Machine learning algorithm	AUC	AP	AUC	AP	AUC	AP	AUC	AP
Logistic regression	0.791	0.693	0.741	0.317	0.650	0.579	0.780	0.328
Random forest	0.941	0.887	0.935	0.486	0.909	0.855	0.931	0.498
Gradient boosting classifier	0.784	0.713	0.797	0.271	0.739	0.650	0.830	0.359
SVM	0.837	0.733	0.782	0.329	0.775	0.708	0.863	0.289
LightGBM	0.964	0.955	0.942	0.848	0.968	0.963	0.966	0.917

ISSN: 1756-0381