Fig. 3

Diagnostic power of DeltaMSI to predict dMMR status at sample level. A representative AUC of logistic regression, SVC and the combined voting model (DeltaMSI) on 28 marker loci in the test set in one of 50 representative bootstrap simulations, as compared to mSINGS on these same 28 loci and same bootstrap simulations (mSINGS) and an optimized, previously published [12] mSINGS done on 10 top-performing marker regions with a fixed optimized baseline set (mSINGS10). B Concordance of classification at sample level (blue dots, dMMR, IHC positive for loss MLH1/MSH2/PMS2 or MSH6 expression; green dots, pMMR, IHC negative) by logistic regression and support vector classifier and provisional dual thresholding with gray zone interval for subsequent clinical validation on real-world data