Table 1.

The performance of different mechanisms at predicting each of the three cancer phenotypes assessed by the Area under the ROC curve (AUC)

Prediction Task	Mechanism	k-TSPs		RF		SVM		XGB
		Train	Test	Train	Test	Train	Test	Train	Test
Bladder cancer progression	FFLsa	0.75	0.73	0.98	0.79	0.91	0.82	0.88	0.78
	NOTCH-MYC	0.85	0.72	1.00	0.70	0.98	0.68	1.0	0.67
	Cell adhesion/O2 response	0.89	0.76	1.00	0.70	1.0	0.64	1.0	0.71
	Alzheimer	0.60	0.49	0.61	0.52	0.54	0.43	0.59	0.52
	Diabetes	0.82	0.65	0.96	0.68	0.86	0.60	0.75	0.52
	Viral infection	0.81	0.63	0.96	0.59	0.90	0.50	0.82	0.57
Response to NACT in TNBC	FFLs	0.74	0.64	0.85	0.73	0.79	0.66	0.75	0.69
	NOTCH-MYCb	0.86	0.75	1.00	0.86	1.0	0.87	1.0	0.87
	Cell adhesion/O2 response	0.88	0.79	1.00	0.85	1.0	0.82	1.0	0.82
	Alzheimer	0.68	0.61	0.77	0.66	0.77	0.59	0.69	0.64
	Diabetes	0.78	0.75	0.94	0.79	0.98	0.74	0.84	0.72
	Viral infection	0.81	0.73	0.96	0.77	0.95	0.69	0.89	0.70
Prostate cancer metastasis	FFLs	0.61	0.63	0.64	0.62	0.62	0.60	0.61	0.60
	NOTCH-MYC	0.73	0.73	0.99	0.69	1.0	0.73	0.83	0.71
	Cell adhesion/O2 responsec	0.80	0.75	0.97	0.73	0.97	0.71	0.85	0.73
	Alzheimer	0.58	0.50	0.59	0.53	0.50	0.50	0.60	0.53
	Diabetes	0.69	0.67	0.89	0.69	0.70	0.68	0.71	0.69
	Viral infection	0.68	0.65	0.81	0.64	0.73	0.63	0.72	0.67

Three important cancer phenotypes were considered for prediction: bladder cancer progression from non-muscle invasive to muscle-invasive stages, response to neoadjuvant chemotherapy (NACT) in patients with triple-negative breast cancer (TNBC), and metastatic progression in prostate cancer. For each phenotype, we built a priori mechanism capturing the underlying biology and used it for prediction in the main analysis. FeedForward Loops (FFLs) were designed for predicting bladder cancer progression while the NOTCH-MYC signaling and cellular adhesion and O₂ response mechanisms were developed for predicting the response to NACT in TNBC and prostate cancer metastatic progression, respectively. Three cancer unrelated mechanisms were used as negative controls: Alzheimer, diabetes, and viral infection. The performance of the different cancer-related and unrelated mechanisms at predicting each of the three cancer phenotypes was assessed using the Area Under the ROC Curve (AUC). FFLs: feedforward loops, NACT: neoadjuvant chemotherapy, TNBC: triple-negative breast cancer.

^aThe feedforward loops mechanism was used for predicting bladder cancer progression in the main analysis.

^bThe NOTCH-MYC signaling mechanism was used for predicting the response to neoadjuvant chemotherapy in triple-negative breast cancer.

^cThe cell-cell adhesion and O2 response mechanism was used for predicting prostate cancer metastasis.