Fig. 3

Precision and recall performance of MuTect and Strelka at each cellularity with different read preprocessing pipelines. The seven data points for each tool are represented as pie charts which correspond to the different cellularities (5 % tumour to 100 % tumour. Novo refers to Novoalign, GATK indicates that collapse, realignment and recalibration was performed, while Raw pipelines only include the collapse (GATK realign and recalibration were skipped). MuTect performs better when BWA is the aligner, and Strelka has high performance in all pipelines. Raw pipelines tend to call more true somatic SNVs, while GATK pipelines report fewer false positives. Note that the Novo-Raw-Strelka and Novo-Raw-MuTect results featured here are identical to those from Phase 1 and Fig. 2