Fig 1.
The input is aligned BAM files from the samples under study, and at least 2 reference normals (5–10 recommended, see methods), as well as liberal variant calls. SuperFreq filters the preliminary SNVs for artefacts using quality scores in the BAM file, and through comparison to the reference normals. Somatic SNVs are called from the remaining variants, while heterozygous germline SNPs are identified for CNA calling. CNAs are identified based on differences in coverage and detecting shifts in allele frequency at heterozygous germline SNPs. Finally, somatic SNVs and CNAs are analysed across samples to designate and track clones.
Table 1.
Properties of other mutation callers and clonal trackers in comparison to SuperFreq.
Fig 2.
Precision and recall of somatic SNV calling across 304 TCGA participants and 33 cancer types.
(A) Recall of somatic SNVs called by the other four callers. (B) Number of unique somatic calls generated by each caller. (C) Recall of coding somatic SNVs from SuperFreq without a matched normal, using SuperFreq cancer-normal analysis as truth. Violins from left: cancer sample alone without filtering on the germlineLike flag, cancer sample alone filtered on the germlineLike flag, cancer sample paired with an in-silico dilution of the cancer and matched normal between 10% and 90%, filtered on the germlineLike flag. (D) Number of false coding SNV calls in the same sample configurations.
Fig 3.
Comparison of somatic CNA calls for 304 TCGA participants across 33 cancer types.
(A) Distribution across participants of the fraction of the genome that agrees with ASCAT. Agreement is defined as LFC and BAF within 0.145 and 0.05 respectively, roughly corresponding to a 20% clonality gain or loss. The blue violins show only participants with a ploidy estimate within 0.2 of ASCAT. (B) SuperFreq recall of somatic CNAs in diluted and sliced samples analysed without the matched normal, with the original cancer-normal analysis used as truth. Each bin is based on at least 100 copy number segments. (C) Comparison of ploidy estimates between ASCAT, ABSOLUTE and SuperFreq, coloured by number of somatic clones called by SuperFreq. Samples were excluded if no somatic clone was called. (D) Recall of gain, loss and CNN-LOH binned by size of the segment, limited to participants where the ploidy agrees within 0.2 between the methods.
Fig 4.
Precision and recall of SuperFreq clonal tracking.
(A) Overview of simulations. As illustrated on the left, the genome is divided into four regions (chr 1–3, 4–8, 9–14 and 15-Y), and the cancer and normal samples are blended to create three samples that contain four clones supported by mutations that reside in different regions of the genome. The expected clonalities across the three samples are shown to the right, where P denotes the purity of the cancer sample. (B) Sensitivity to find clones with a matched normal (WMN) or no matched normal (NMN) in SuperFreq and SciClone, as function of maximum clonality. (C) Recall of the four simulated clones, binned on the purity of the original cancer sample. (D) Number of false clones called. (E) Fraction of mutations associated with a clone originating from the expected chromosomes in panel A. (F) Fraction of mutations called in the cancer-normal recalled by the called clones.
Fig 5.
Assessment of clonal structure in AML using SuperFreq.
(A) Clone scatter plots of four AML patients. The VAF is shown for all somatic point mutations that are tracked by SuperFreq, coloured by the clone they are assigned to. Selected mutations are highlighted, including CNAs that are assigned a VAF of half of the clonality. Point size is used to indicate read depth, with larger points reflecting higher coverage. River plots and line plots showing the progression of the clones across the Diagnosis (Dx), sorted normal (N) and relapse (Rel) samples for (B) AML.084 and (C) AML.110. Key mutations are highlighted in the line plots.
Fig 6.
Assessment of clonal structure in AML using SuperFreq without matched normal controls.
Clone scatter plots of four AML patients (as in Fig 5), generated without using the matched normal sample. We include the germline clone to illustrate how it absorbs germline variants, as well as some somatic variants that have high clonality in both samples. While this is done successfully in AML.080 and AML.110, the two other samples have a large number of variants designated as somatic that are not identified in the cancer-normal analysis. AML.084 and AML.110 identify clones at low clonality in both samples that are not seen in the cancer-normal analysis, shown in grey.