Can someone explain these purity plots produced by PURPLE? Is there any documentation on interpreting them? The top sample has a purity of 62%, fit method 'somatic' and QC passed. The below sample has a purity of 8%, 'no tumour' and QC failed. Why is the relative score higher for my sample that seems to be poorer quality?

Hi Cathal, could you share the circos as well by any chance? What kind of sample is this ?

Hi @Rayan Hassaïne These samples are brain tumours from FFPE tissues. Here are the circos plots next to the purity plots for both samples. The sample with 62% has chromosomal rearrangements and the other sample does not. However the quality pass/fail is different for both too. Why is a lower score better? Can I say that the sample with 62% has more tumour present? Does it mean that 62% of the sample is approx tumour?

@Cathal the detailed documentation is on our GitHub: https://github.com/hartwigmedical/hmftools/blob/master/purple/README.md#4-sample-purity-and-ploidy

Hi @Joep de Ligt and @Rayan Hassaïne I understand that tumour purity does not always correspond to quality. I have read the github page but it seems to relate more to the background theory rather than simple interpretation of the outputs with examples. I am interested in finding out how the circos plots and tumour purity plots are related. i.e. what does it mean if for a sample has expected copy number alterations can but has low tumour purity, can that sample be analysed further? We are also interested in choosing a subset of samples to send for spatial transcriptomics analysis, what metrics can be looked at to choose the best samples? If a sample has low tumour purity then is it reasonable to assume that the FFPE block would be of lower quality overall. Also in relation to relative scores, can I compare 2 purity plots (as in example above) and say that this sample has more than the other? Or are those plots independent.

Hi @Cathal good to know that you've read the theory, from what you're saying there a couple of questions I can distill that I'll try to answer. 1. You have some prior (orthogonal?) knowledge of these tumors and you're wanting to understand how the OA output relates to this. Please note that we only report (diagnostically) on patients with a sample that has at least 20% tumor content so the first sample would be processed and assessed for it's further quality (which would likely be issued with strong QC warnings), the second sample would be failed. a. the CopyNumber noise you see in Circos for the second sample is noise not a real signal, most likely due to the fact that 1. there is almost nu tumor detected, 2 the software tries to still fit a purity/ploidy and based on that call events, but this will be very unreliable and the CNs called are most likley FP due to FFPE fragmentation not biology b. the first sample shows CN signal but the tool is very uncertain about what exactly the best fit would be, i.e. all the blue areas are possible solutions to the purity ploidy fit problem. The example from the docs is maybe easier to read (see attached). Here you can see 1 optimal solution around ploidy 3 an a 'light green' possible but less likely fit at ploidy 1.8. The fact you see so much blue means 2 things, 1. it's a noisy sample (again likely due to FFPE. 2. there are multiple equally likely solutions, if you have prior knowledge (FISH, staining, etc, you can restrict the search space to what you think is most likely 2. You want to possibly select samples for downstream assays, this really depends on wether the FFPE damage that has been introduced is an issue. I.e. both samples clearly suffer from it in their quality so that could be a reason to exclude both. If it's purely tumor content you're worried about you could consider taking the first sample along.

That makes sense, thanks @Joep de Ligt