Learning models for colorectal cancer signature reconstruction and classification in patients with chronic inflammatory bowel disease

Table 1 Possible comparisons to be made during the unsupervised (i.e., we do not rely on the type of disease) global analysis of patients following the considered three different strategies

Strategy	Comparison
Strategy 1 (classic)	1 vs (2,3,4)
	2 vs (3,4)
	3 vs 4
Strategy 2 (1&1)	1 vs 2; 1 vs 3; 1 vs 4
	2 vs 1; 2 vs 3; 2 vs 4
	3 vs 1; 3 vs 2; 3 vs 4
	4 vs 1; 4 vs 2; 4 vs 3
Strategy 3 (pairwise)	1 vs (2,3,4)
	2 vs (1,3,4)
	3 vs (1,2,4)
	4 vs (1,2,3)

Table 2 Summary of patients’ classification predicted by random forests/support vector machines respectively. From left to right: Group of patients, amount of selected miRNA, percentage of success in true positive classification, sensitivity, specificity and their area under the curve

Methods	Nº miRNA	% True classification (95%CI)	Sensitivity	Specificity	AUC
All miRNA
Strategy 1	56	69 (62-75)/69 (62-75)	0.25/0.43	0.93/0.83	0.76/0.74
CD	9	87 (78-93)/86 (77-92)	0.70/0.73	0.96/0.93	0.89/0.92
UC	30	72% (63-80)/76 (67-83)	0.45/0.55	0.86/0.87	0.77/0.81
miRNAs selected by sPLS-DA
Strategy 1	11	69 (62-75)/68 (62-75)	0.36/0.36	0.87/0.86	0.72/0.74
CD	5	80 (70-88)/82 (67-86)	0.67/0.60	0.87/0.87	0.84/0.86
UC	8	73 (64-80)/81 (73-88)	0.48/0.57	0.86/0.93	0.73/0.81

AUC: Area under curve; CD: Crohn’s disease; UC: Ulcerative colitis.

Table 3 All patients contingence matrix of the 56-selected miRNAs by means of random forests and support vector machines methods

Predicted by RF Predicted by SVM
		Cases	Controls	Cases	Controls
True	Case	18	54	31	41
	Controls	10	124	23	111

RF: Random forests; SVM: Support vector machines.

Table 4 Contingence matrix of the 9-selected miRNA and random forests methods for Crohn’s disease patients

Predicted by RF Predicted by SVM
		Cases	Controls	Cases	Controls
True	Case	21	9	22	4
	Controls	2	53	8	51

RF: Random forests; SVM: Support vector machines.

Table 5 Contingence matrix of the 30-selected miRNA and random forests methods for Ulcerative colitis patients

Predicted by RF Predicted by SVM
		Cases	Controls	Cases	Controls
True	Case	19	23	23	19
	Controls	11	68	10	69

RF: Random forests; SVM: Support vector machines.

Table 6 Contingence matrix of the 11-selected miRNA and random forests methods for all patients

Predicted by RF Predicted by SVM
		Cases	Controls	Cases	Controls
True	Case	27	45	26	46
	Controls	18	116	19	115

RF: Random forests; SVM: Support vector machines.

Table 7 Contingence matrix of the 5-selected miRNA and random forests methods for Crohn’s disease patients

Predicted by RF Predicted by SVM
		Cases	Controls	Cases	Controls
True	Case	20	10	20	10
	Controls	7	48	5	50

RF: Random forests; SVM: Support vector machines.

Table 8 Contingence matrix of the 9-selected miRNA and random forests methods for Ulcerative colitis patients.

Predicted by RF Predicted by SVM
		Cases	Controls	Cases	Controls
True	Case	20	22	24	18
	Controls	11	68	5	74

RF: Random forests; SVM: Support vector machines.