TABLE 7.

Performance results on the external validation data.

Experiments using different gene levels (1–5–30–50)	Sensitivity	Specificity	Accuracy	F-measure
Random 1 gene	0.43	0.58	0.51	0.44
Top 1 gene of mirModuleNet	0.84	0.88	0.87	0.85
Random 5 genes	0.46	0.61	0.55	0.48
Top 5 genes of mirModuleNet	0.94	0.81	0.88	0.87
Random 30 genes	0.57	0.91	0.76	0.68
Top 30 genes of mirModuleNet	0.94	0.92	0.93	0.92
Random 50 genes	0.76	0.94	0.86	0.83
Top 50 genes of mirModuleNet	0.94	0.97	0.95	0.94

In all experiments, the model is trained on TCGA- LUSC data and tested on external data, which is LUSC_E.