Precision of raw XHMM and DECoNT-corrected XHMM calls are shown with respect to varied confidence threshold (DEL, DUP, and overall precision); 6832 CNV calls made by XHMM on our test data set (the 1000 Genomes Project WES data set test samples) are used. The ground truth is the CNV calls made by CNVnator on the corresponding WGS samples. The last panel shows the number of calls remaining as the threshold gets more conservative. For each polished tool, we used 90% of the calls made on 802 1000 Genomes Project samples for training and the remaining 10% of the calls for testing. This roughly corresponds to a test set size of 80 samples.
