Whole-genome doubling-aware copy number phylogenies for cancer evolution
Project description
MEDICC2 - Whole-genome doubling-aware copy number phylogenies for cancer evolution
For more information see the accompanying paper Whole-genome doubling-aware copy number phylogenies for cancer evolution with MEDICC2.
Installation
Install MEDICC2 via conda (recommended), pip or from source. MEDICC2 was developed and tested on unix-built systems (Linux and MacOS). For Windows users we recommended WSL2.
Note that the notebooks and examples are not included when installing from conda or pip. When installing from pip or source, you need to make sure to have a working version of gcc
and gxx
installed.
Due to dependencies we recommend using Python version 3.7-3.9.
Installation via conda (recommended)
MEDICC2 can be installed via conda install -c bioconda -c conda-forge medicc2
.
Installation via pip
As MEDICC2 relies on OpenFST version 1.8.1 which is not packaged on PyPi you have to first install it using conda with conda install -c conda-forge openfst=1.8.1
. Next you can install MEDICC2 via pip install medicc2
.
Installation from source
Clone the MEDICC2 repository and its submodules using git clone --recursive https://bitbucket.org/schwarzlab/medicc2.git
. It is important to use the --recursive
flag to also download the modified OpenFST submodule.
All dependencies including OpenFST (v1.8.1) should be directly installable via conda. A yaml file with a suggested MEDICC2 conda environment is provided in 'doc/medicc2.yml'. You can create a new conda environment with all requirements using conda env create -f doc/medicc2.yml -n medicc_env
.
Then, inside the medicc2
folder, run pip install .
to install MEDICC2 to your environment.
Usage
After installing MEDICC2, you can use MEDICC2 functions in python scripts (through import medicc
) and from the command line. General usage from the command line is medicc2 path/to/input/file path/to/output/folder
. Run medicc2 --help
for information on optional arguments.
Logging settings can be changed using the medicc/logging_conf.yaml
file with the standard python logging syntax.
Command line Flags
input_file
: path to the input fileoutput_dir
: path to the output folder--input-type
,-i
: Choose the type of input: f for FASTA, t for TSV. Default: 'TSV'--input-allele-columns
,-a
: Name of the CN columns (comma separated) if using TSV input format. This also adjusts the number of alleles considered (min. 1, max. 2). Default: 'cn_a, cn_b'--input-chr-separator
: Character used to separate chromosomes in the input data (condensed FASTA only). Default: 'X'--tree
: Do not reconstruct tree, use provided tree instead (in newick format) and only perform ancestral reconstruction. Default: None--topology-only
,-s
: Output only tree topology, without reconstructing ancestors. Default: False--normal-name
,-n
: ID of the sample to be treated as the normal sample. Trees are rooted at this sample for ancestral reconstruction. If the sample ID is not found, an artificial normal sample of the same name is created with CN states = 1 for each allele. Default: 'diploid'--exclude-samples
,-x
: Comma separated list of sample IDs to exclude. Default: None--filter-segment-length
: Removes segments that are smaller than specified length. Default: None--bootstrap-method
: Bootstrap method. Has to be either 'chr-wise' or 'segment-wise'. Default: 'chr-wise'--bootstrap-nr
: Number of bootstrap runs to perform. Default: None--prefix
, '-p': Output prefix to be used. None uses input filename. Default: None--no-wgd
: Disable whole-genome doubling events. Default: False--plot
: Type of copy-number plot to save. 'bars' is recommended for <50 samples, heatmap for more samples, 'auto' will decide based on the number of samples, 'both' will plot both and 'none' will plot neither. (default: auto).--total-copy-numbers
: Run for total copy number data instead of allele-specific data. Default: False-j
,--n-cores
: Number of cores to run on. Default: None--chromosomes-bed
: BED file for chromosome regions to compare copy-number events to--regions-bed
: BED file for regions of interest to compare copy-number events to-v
,--verbose
: Enable verbose output. Default: False-vv
,--debug
: Enable more verbose output Default: False--maxcn
: Expert option: maximum CN at which the input is capped. Does not change FST. Default: 8--prune-weight
: Expert option: Prune weight in ancestor reconstruction. Values >0 might result in more accurate ancestors but will require more time and memory. Default: 0--fst
: Expert option: path to an alternative FST. Default: None--fst-chr-separator
: Expert option: character used to separate chromosomes in the FST. Default: 'X'
Input files
Input files can be either in fasta or tsv format:
- fasta: A description file should be provided to MEDICC. This file should include one line per file with the name of the chromosome and the corresponding file names. If fasta files are provided you have to use the flag
--input-type fasta
. - tsv: Files should have the following columns:
sample_id
,chrom
,start
,end
as well as columns for the copy numbers. MEDICC expects the copy number columns to be calledcn_a
andcn_b
. Using the flag--input-allele-columns
you can set your own copy number columns. If you want to use total copy numbers, make sure to use the flag--total-copy-numbers
. Important: MEDICC2 does not create total copy numbers for you. You will have to calculate total copy numbers yourself and then specify the column using the--input-allele-columns
flag.
MEDICC2 follows the BED convention for segment coordinates, i.e. segment start is at 0 and the segment end is non-inclusive.
The folder examples/simple_example
contains a simple example input both in fasta and tsv format.
The folder examples/OV03-04
contains a larger example consisting of multiple fasta files. If you want to run MEDICC on this data run medicc2 examples/OV03-04/OV03-04_descr.txt path/to/output/folder --input-type fasta
.
Output files
MEDICC creates the following output files:
_final_tree.new
,_final_tree.xml
,_final_tree.png
: The final phylogenetic tree in Newick and XML format as well as an image_pairwise_distances.tsv
: A NxN matrix (N being the number of samples) of pairwise distances calculated with the symmetric MEDICC2 distance_final_cn_profiles.tsv
: Copy-number profiles of the input as well as the newly internal nodes. Also includes additional information such as whether a gain or loss has happened_copynumber_events_df.tsv
: List of all copy-number events detected. Note that entries for WGD events have non-meaningful values for chrom, cn_child, ..._cn_profiles.pdf
: Combined plot of the phylogenetic tree as well as the copy-number profiles of all samples (including the internal nodes)_events_overlap.tsv
: Overlap of copy-number events with regions of interest (see below)
Output plots
Apart from the file _tree.pdf
which contains the inferred phylogeny, the main plot created by MEDICC is the copy-number plots named either _cn_profiles.pdf
or _cn_profiles_heatmap.pdf
.
The left part consists of the inferred phylogenetic tree including the number of events in the branches. The right part is made up of the copy-number profiles of the samples (and potentially the reconstructed ancestral nodes).
There are two kinds of copy-number plots: the bars and the heatmap version. The bars version is most suitable for fewer samples (<50) as more details are visible while the heatmap version is most suitable many samples expected for example in single-cell experiments.
You can toggle the kind of plot MEDICC2 creates with the --plot
flag (see above).
Example bars copy-number plot
Example from patient PTX011 from the Gundem et al. Nature 2015. The data can be found in example/gundem_et_al_2015/
.
Legend
Example heatmap copy-number plot
Example from patient PTX011 from the Gundem et al. Nature 2015. The data can be found in example/gundem_et_al_2015/
.
Usage examples
For first time users we recommend to have a look at examples/simple_example
to get an idea of how input data should look like. Then run medicc2 examples/simple_example/simple_example.tsv path/to/output/folder
as an example of a standard MEDICC run. Finally, the notebook notebooks/example_workflows.py
shows how the individual functions in the workflow are used.
The notebook notebooks/bootstrap_demo.py
demonstrates how to use the bootstrapping routine and notebooks/plot_demo.py
shows how to use the main plotting functions.
Regions of interest
MEDICC2 compares the detected copy-number events to regions of interest. These regions are chromosome-boundaries and known oncogenes and tumor-suppressor genes. By default MEDICC2 uses hg38 chromosome-arms and a list of genes taken from Davoli et al. Cell 2013. This data is present as BED files in the medicc/objects
folder.
Users can specify regions of interest of their own in BED format by providing the --chromosomes-bed
or --regions-bed
flags.
Issues
If you experience problems with MEDICC2 please file an issue directly on Bitbucket or contact us directly.
Known Issues
Noisy segments
Small faulty or noisy segments can have a strong effect on the distances MEDICC2 calculates between samples and therefore the resulting tree.
This is because MEDICC2 counts all segments equally in order appropriatlely take focal events into account.
If the resulting and the inferred events look strange to you, you can replot the tree and copy-number profiles using the function plot_cn_profiles
setting ignore_segment_lengths=True
(see the notebook notebooks/plot_demo.py
for usage examples) in order to investigate small segments that might not have been visible in the original plot.
If you are unsure about the copy-number profiles we recommened to filter small segments.
Taxon imbalance If your data contains 100s to 1000s samples with a few distinct subgroups, an imbalance in the number of samples per subgroups might lead to an incorrect tree (e.g. 50 samples of subclone A and 1000 samples each of subclone B and C). This is a known problem in phylogeny called taxon imbalance or taxon sampling. If you have multiple, clearly separable subgroups in your data we recommoned either subsampling over-represented groups or upsampling under-represented groups to gauge the effect of taxon imbalance.
Running out of memory / bad_alloc error
If MEDICC2 terminates with the following error terminate called after throwing an instance of 'std::bad_alloc'
or your machine runs out of memory this hints towards an issue with the FST.
Rerun MEDICC2 with the -vv
flag to enable extended logging. If the error occurs during the ancestral reconstruction routine, the issue is related to OpenFST which is the FST library employed by MEDICC2 and cannot be easily solved by us.
This issue can be related to small bin sizes (and therefore a large number of segments). Increasing the binsize (although decreasing accuracy) solves this issue most of the time.
You can also try to remove the sample that led to the error (see the extended logs for this).
The output plots are not like I expected
Maybe you need to set the --plot
flag. By default, --plot
is set to auto which means that it plots different figures depending on the number of samples in the data (threshold is 50); see above.
Contact
Email questions, feature requests and bug reports to Tom Kaufmann, tom.kaufmann@mdc-berlin.de.
License
MEDICC2 is available under GPLv3. It contains modified code of the pywrapfst Python module from OpenFST as permitted by the Apache 2 license.
Please cite
Kaufmann TL, Petkovic M, Watkins TBK, Colliver EC, Laskina S, Thapa N, Minussi DC, Navin N, Swanton C, Van Loo P, Haase K, Tarabichi M, Schwarz RF.
MEDICC2: whole-genome doubling aware copy-number phylogenies for cancer evolution
bioRxiv 2021 Sep 6; doi: 10.1101/2021.02.28.433227
Schwarz RF, Trinh A, Sipos B, Brenton JD, Goldman N, Markowetz F.
Phylogenetic quantification of intra-tumour heterogeneity.
PLoS Comput Biol. 2014 Apr 17;10(4):e1003535. doi: 10.1371/journal.pcbi.1003535.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.