1. FAQ
REVOLVER is an R tool to:
- compute phylogenetic trees for a cohort of patients. For each patient the tools can process one or more (i.e., multi-region) samples;
- fit correlated trees across patients; the correlation is measured by extracting evolutionary trajectory from driver alterations annotated in the data. These drivers shall occur in multiple patients, so that the trajectories are repeated across patients;
- split the cohort in groups of tumours that are shaped by similar evolutionary pressures, after the fit.
- plot data, models, clusters, confidence etc.
The details of the method are described in the main paper.
REVOLVER can process both Cancer Cell Fractions (CCF) and binary data, but only one type of data at a time (you should not use CCF for some patients with, and binary data for others).
The image above summarises the two options (CCF and binary data), after you have called somatic mutations and copy number across all your cohort.
-
CCF These are computed from mutations and copy number calls and report the proportion of cancer cells where they occur; for instance, mutations with CCF ~= 1.0 are clonal. You need to compute CCFs for you patients before REVOLVER; you can do that by using tools for sub-clonal deconvolution that estimate the subpopulations admixture in your bulk samples. This is a common step for a phylogenetic analysis from bulk sequencing, you can use several tools like pyClone, sciClone etc. Feel free to drop as a line if you have questions.
-
If you use binary data, you can immediately use REVOLVER. This type of data has lower-resolution compared to CCF.
Yes, the tool computes phylogenetic trees from CCFs, or mutation trees from binary data. If however you prefer to compute the trees with any other tool you can import them in the tool..
There are no hard constraint, but rather suggestions that apply to any "Cancer Evolution" analysis. For instance, to use CCFs and phylogenetic reconstruction, whole-genome sequencing is the best data to deconvolve cancer subpopulations. However, you can carry out a CCF-based analysis from whole-exome sequencing, as shown in the TRACERx study that we analysed with REVOLVER. To use just binary data and mutation trees, you can also use a simpler targeted panel.
It is hard to give minimum values for these parameter, as they change with each tumour type and tissue. In general, both impact on the quality of the analysis; if you want to detect small sub-clones, or low-frequency mutations in a CCF-based analysis, you want to sequence at high coverage samples with "good" purity. The TRACERx study, for instance, reached ~430x median coverage on exome regions.
This depends on the level of spatial heterogeneity of the tumour under scrutiny. The framework supports any number of samples and, in principle, with CCFs, you can also use single-samples datasets. With binary data, you need multi-region sampling. So, you can run REVOLVER on large-scale studies like TCGA only with CCF. However, the extent to which a single-biopsy can capture the phylogenetic structure of a complex tumour is debatable.
Because you want to look for repeated evolutionary trajectories, you want a large number of patients to detect reliable statistical signals from the input data. The best cohorts that we analyzed so far, which are the largest available to date, have either 50 or 100 patients each.
We did not focus the manuscript on this topic because there are not yet many multi-patient single-cell datasets available in the community. The statistical framework, however, and the implementation support this type of data. This is because once we have built the trees for each patient, then the inference is independent of the "type of data" that defines the trees. Thus, since single-cell data is binary, you will use REVOLVER's mutation trees that employ an infinite site assumption, or your own tree-generation method.
If you have any other question or you need support, drop us a line.
Imagine that we detect, in our patient PATIENT, two or more driver mutations in different positions/ nucleotides of a gene GENE: in a phylogenetic analysis, these driver mutations would appear in (possibly) different positions of the output tree. It is possible to handle these events also in REVOLVER, but this requires some considerations. The method computes correlated evolutionary trajectories among recurrent drivers. To extract and correlate them, REVOLVER uses an ID-based identification system: in many cases this could be just the gene name GENE. The ID is important, as it is used as a key to "identify" the driver and to see in which patients it occurs: thus the ID has to be unique. In the above case, calling all the drivers GENE is bad choice because the ID is not unique: we would have multiple entries in PATIENT with ID GENE, and this will break the computation. We can circumvent this by using a finer-grained resolution in the preparation of our data: we can add to the ID the nucleotide position of the mutation, for instance: i.e., something like GENE_XXX and GENE_YYY, where XXX and YYY are positions/ domains would work. Doing so the driver events would be correlated independently, and everything would work fine. A drawback could be, however, that those drivers might not be any more "recurrent" across the cohort: because GENE_XXX and GENE_YYY are different "events", they would be recurrent only if we have other patients with the same driver mutations (i.e., annotated in the same gene loci). This can create a problem if we do not have a large cohorts, of course. It is always possible, of course, to keep in the data only one of the driver mutations on the gene, and annotate it with ID GENE. This comes at the expense of neglecting these parallel evolution events. As far as we could measure in large cohorts, however, these events are not so frequent to become a hurdle to our analysis.
REVOLVER's input requires also a "group" assignment for each one of the annotated mutations. The definition of groups depends on the data used:
- CCF data. we perform sub-clonal deconvolution to assign mutations to the clones in our tumour, prior to tree reconstruction. This is done by clustering read counts from input data. In this case, we can use clone assignments to define the groups.
- Binary data. With mutation trees from binary data, we do not carry out sub-clonal deconvolution. In this case, a groups is just defined by the set of mutations that occur in the same set of samples. In this case, each group is not a proper "clone", as if we were using CCF data.
A phylogenetic tree is a tree built from CCFs, real values that provide the estimate of cancer cells harbouring a certain SNV/ Copy Number/ etc. A mutation tree is built from binary data, which is a simple 0/1 format to represent the absence/ presence of a certain SNV/ Copy Number/ etc. in a sample. In both cases, the tree is build over the groups annotated in the data.