Table 1. MSC Signature improves the classification accuracy of MSC compared to a panel of 32 commonly used MSC markers.

Column 1 provides the comparison of the classification accuracy of the 635 training samples using (Column 2) the three markers used as the minimal immunophenotype of the MSC training samples. (Column 3) a panel of 32 commonly used immune-markers in the MSC literature; (Column 4) using the unrefined sPLS-DA output; or (Column 5) with our final signature of 16 genes. Performance of each gene group was assessed using 200 random subsamplings of the training set. The internal classification error rate was calculated from a PLS-DA with 2 components (known immune-markers), or was an output of our statistical model with genes selected in an unbiased manner (cf Fig. 1A).

	CD45, CD73, CD105	32 common MSC markers	sPLS-DA prior to stable gene selection	The 16-gene MSC signature
Overall accuracy (% of 635 samples)	87.86	92.33	97.71	97.85
MSC misclassified (% of 125 samples)	14.40	11.10	3.04	4.31
Non-MSC misclassified (% of 510 samples)	11.60	6.82	2.11	1.61