Fig. 4

Results on real sequencing data. Sequence reads originating from a single lane of an Illumina HiSeq 2000 run of NA12878 were aligned by BWA and Bowtie 2 to the human genome and the alignments were filtered to the target region of the whole exome. Variants were called by GATK HaplotypeCaller, GATK UnifiedGenotyper, FreeBayes and SAMtools and the unfiltered call sets were combined by VariantMetaCaller and BAYSIC. Each variant call sets were hard filtered according to general recommendations and the GATK-based variant calls were also recalibrated by VQSR. All call sets were filtered to the confidently called region of the Platinum reference call set. a Precision-recall curves for each method for SNPs (top) and for indels (bottom) in case of BWA (left) and Bowtie 2 (right) alignment. b Area under the precision-recall curves for each method for SNPs (top) and for indels (bottom) in case of BWA (left) and Bowtie 2 (right) alignment. c Mean absolute error of estimated versus true precision of the different methods that produce probability estimates of variants. FB = FreeBayes, HC = HaplotypeCaller, HF = Hard filtered, MAE = mean absolute error, ST = SAMtools, UG = UnifiedGenotyper, VMC = VariantMetaCaller, VQSR = Variant quality score recalibration