Table 2.

Differences in predictive accuracy related to certain versus uncertain sample predictions in the training set (95 biopsy samples, 25 patients).

Uncertainty threshold	Number of samples (n)	Overall accuracy	Sensitivity/specificity (%/%)
Entire pooled cohort	95	75%	77/74
p < 0.05 (certain)	72	83%	83/83
p > 0.05 (uncertain)	23	48%	63/40
p < 0.10 (certain)	78	79%	83/78
p > 0.10 (uncertain)	17	53%	63/44
p < 0.15 (certain)	81	79%	84/77
p > 0.15 (uncertain)	14	50%	57/43

Shown are the differences in predictive performance of the transductive learning GP model for the entire pooled training cohort and also when stratifying the sample predictions with high certainty (low p values) versus low certainty (high p values) at various p value thresholds.