SuperFreq: Integrated mutation detection and clonal tracking in cancer biorxiv.org/content/early/…
If you want to use: github.com/ChristofferFle…
I've developed and used this the last 4+ years, and now finally out! 🎉
Let me make a thread (or tree?) from here.
If you want to use: github.com/ChristofferFle…
I've developed and used this the last 4+ years, and now finally out! 🎉
Let me make a thread (or tree?) from here.
What is superFreq?
It analyses cancer exomes, especially geared for multiple matched cancer samples.
- calls somatic SNVs
- calls CNAs
- calls clones and phylogeny
as well as quality of life:
- annotates variants (effect, dbSNP, ExAC, COSMIC)
- lots of plots: results and QC
It analyses cancer exomes, especially geared for multiple matched cancer samples.
- calls somatic SNVs
- calls CNAs
- calls clones and phylogeny
as well as quality of life:
- annotates variants (effect, dbSNP, ExAC, COSMIC)
- lots of plots: results and QC
Why should you use it?
Not because it beats some other method in precision/recall on simulated data, but because it's been developed to answer questions from ours and others data. So focus has been on:
- robust to noise
- working without matched normal
- lots of QC
Not because it beats some other method in precision/recall on simulated data, but because it's been developed to answer questions from ours and others data. So focus has been on:
- robust to noise
- working without matched normal
- lots of QC
My favourite plot! It compares somatic SNV calls from 5 methods across 300 cancers. 5 way venn diagram? 300 of those? Not a great idea...
What if I told you that you can get a good overview of that in a single plot?
What if I told you that you can get a good overview of that in a single plot?
We really only barely mention this in the paper, but it's pretty cool imo, and I'm not sure anyone else does this in an automated way. When multiple matched samples are analysed, superFreq checks if it's the same allele that's affected by each CNA, and acts accordingly.
As most of superFreq, it's inspired by an analysis of real patient data, in this case from biorxiv.org/content/early/…. Especially Supplementary figure 4. We could split a gain of chr7 into different events, pointing to convergent evolution.
I looked through the river plots of an AML cohort with diagnosis and relapse sample, and found this, which is how it is shown in superFreq output. chr8 has a AAB genotype call in the orange clone, ABB in the red. Both gain, but of different alleles.
The signal for these is really super strong (if region decently big). This sample a bit noisy, but there is a clean gain of chr8 in both samples. Scattering the VAFs of diagnosis vs relapse on chr8 shows clear off-diagonal clusters, which is what the call is based on.
All the above plots come default from a superFreq run. With infinite time it's be interesting to look at performance of these calls over cohorts and some orthogonal verification, but well. Data from here:
