Table 1.
Design of mixing experiments.
Figure 1.
Circos plots of the SNP array data for a paired normal (ND) and tumor (TD) sample showing regions of LOH in the tumor sample (A). The chromosome ideograms are shown on the outer wheel, the logR and BAF values are plotted in the middle and inner wheel respectively. The density plot of the probes in LOH regions (B) is used to calculate the d-score (C). The d-score is compared to the density plots of probes within regions of LOH for the cell line: normal DNA mixtures which represent different cellularity (D). The d-score and cellularity are highly correlated (E). Three plots from the left to the right are the scatter plot only, with fitting the simple linear model and with fitting the spline regression model respectively.
Figure 2.
B allele frequency (BAF) and log R ratio (LRR) plots for a region of LOH with changing tumor cellularity.
DNA from a cancer cell line and matched normal DNA were mixed in different proportions and assayed using SNP arrays. BAF and LRR plots were generated using GenomeStudio software (Illumina). For illustrative purposes a region of loss on the p arm of chromosome 7 in the cancer cell line is shown. In the 100 normal sample (0
tumor) the SNPs are either heterozygous (BAF
0.5) or homozygous (BAF = 0 or 1). In regions of single chromosome loss in the tumour there is LOH. In the 100
cell line the BAF is showing a homozygous state and there is clear loss in the LRR. As tumour cellularity decreases the separation of the BAF decreases.
Table 2.
The leave-one-out cross-validation results for each model in the qpure method.
Figure 3.
Correlations of cellularity estimated by different methods in a pancreatic cancer cohort.
Cellularity was predicted in the pancreatic cohort using 3 methods: pathology review, qpure and deep Ion Torrent sequencing of KRAS. Cellularity predictions are shown in the boxplot (A), the -value was calculated using an ANOVA test to determine whether on average there is difference between the cellularity scores returned by the different methods. The correlation between each method using Spearman’s rank correlation was calculated (B–D). Scatter plots are shown which compare KRAS deep sequencing and qpure estimates (B), qpure and pathology estimates (C), and KRAS deep sequencing and pathology estimates (D).