Cortex on steroids: for larger populations with better assembly
Multi-sample de novo assembly and variant calling using de bruijn graphs. Variant calling with and without a reference genome. Between closely related samples or highly diverged ones. From bacterial to mammalian genomes. Minimal configuration. And it's free.
Isaac Turner's experimental rewrite of cortex_var, to handle larger populations with better genome assembly. PhD supervisor: Prof Gil McVean. Collaborators: Zam Iqbal, Kiran Garimella. Based at the Wellcome Trust Centre for Human Genetics, University of Oxford.
Note: Currently under development. Expect bugs, fixes and vague documentation until we hit our first release. Feel free to try out McCortex and watch this space for the release.
9 Nov 2015
Branch | Status |
---|---|
master: | |
develop: | |
code analysis: |
McCortex compiles with clang and gcc. Tested on Mac OS X and linux. Requires zlib. Download with:
git clone --recursive https://github.com/mcveanlab/mccortex
To compile for a maximum kmer size of 31:
make all
to compile for a maximum kmer size of 63:
make MAXK=63 all
Executables appear in the bin/
directory.
Download and compile McCortex. Can be in any directory, later I'll assume it's in ~/mccortex/
:
git clone --recursive https://github.com/mcveanlab/mccortex
cd mccortex
make all MAXK=31
make all MAXK=63
Now write a file detailing your samples and their data. Columns are separated by one or more spaces/tabs. File entries are separated by commas. Paired-end read files are separated by a colon ':'. File paths can be relative to the current directory or absolute. Most fileformats are supported:
cd /path/to/your/data
echo "#sample_name SE_files PE_files interleaved_files" > samples.txt
echo "Mickey a.fa,b.fa reads.1.fq.gz:reads.2.fq.gz ." >> samples.txt
echo "Minney . reads.1.fq.gz:reads.2.fq.gz in.bam" >> samples.txt
echo "Pluto seq.fq . pluto.cram" >> samples.txt
Create a job file from your sample file (samples.txt
). All output will go into the directory we specify (mc_calls
). We also specify the kmer(s) to use. We'll run at k=31
and k=61
and merge the results.
If your data are haploid, we set --ploidy 1
:
~/mccortex/scripts/make-pipeline.pl -r /path/to/ref.fa --ploidy 1 31,61 mc_calls samples.txt > job.k31.k61.mk
If your samples are human, you have a mix of haploid and diploid chromosomes. Therefore you need to specify which samples have only one copy of chrX
and one of chrY
. The format is -P <sample>:<chr>:<ploidy>
where <sample>
and <chr>
can be comma-separated lists. Ploidy arguments are read in order.
~/mccortex/scripts/make-pipeline.pl -r /path/to/ref.fa --ploidy "-P .:.:2 -P .:chrY:1 -P Mickey:chrX:1" 31,61 mc_calls samples.txt > job.k31.k61.mk
Now you're ready to run. You'll need to pass:
CTXDIR=
MEM=
(2GB for ten E. coli, 100GB for a human)NTHREADS=
Run the job file:
make -f job.k31.k61.mk CTXDIR=~/mccortex MEM=100GB NTHREADS=8 \
JOINT_CALLING=yes USE_LINKS=no brk-geno-vcf
For a human, running time will be about 8 hours for a single sample and use about 100GB of RAM.
Job finished? Your results are in: mc_calls/vcfs/breakpoints.joint.plain.k31.k61.geno.vcf.gz
.
Something go wrong? Take a look at the log file of the last command that ran. You may need to increase memory or compile for a different MAXK=
value. Once you've fixed the issue, just rerun the make -f job...
command. Add --dry-run
to the make
command to see which commands are going to be run without running them.
usage: mccortex31 <command> [options] <args>
version: ctx=XXXX zlib=1.2.5 htslib=1.2.1 ASSERTS=ON hash=Lookup3 CHECKS=ON k=3..31
Commands: breakpoints use a trusted assembled genome to call large events
bubbles find bubbles in graph which are potential variants
build construct cortex graph from FASTA/FASTQ/BAM
calls2vcf convert bubble/breakpoint calls to VCF
check load and check graph (.ctx) and path (.ctp) files
clean clean errors from a graph
contigs assemble contigs for a sample
correct error correct reads
coverage print contig coverage
dist make colour kmer distance matrix
index index a sorted cortex graph file
inferedges infer graph edges between kmers before calling `thread`
join combine graphs, filter graph intersections
links clean and plot link files (.ctp)
pjoin merge path files (.ctp)
popbubbles pop bubbles in the population graph
pview text view of a cortex path file (.ctp)
reads filter reads against a graph
rmsubstr reduce set of strings to remove substrings
server interactively query the graph
sort sort the kmers in a graph file
subgraph filter a subgraph using seed kmers
thread thread reads through cleaned graph to make links
uniqkmers generate random unique kmers
unitigs pull out unitigs in FASTA, DOT or GFA format
vcfcov coverage of a VCF against cortex graphs
vcfgeno genotype a VCF after running vcfcov
view text view of a cortex graph file (.ctx)
Type a command with no arguments to see help.
Common Options:
-h, --help Help message
-q, --quiet Silence status output normally printed to STDERR
-f, --force Overwrite output files if they already exist
-m, --memory <M> Memory e.g. 1GB [default: 1GB]
-n, --nkmers <H> Hash entries [default: 4M, ~4 million]
-t, --threads <T> Limit on proccessing threads [default: 2]
-o, --out <file> Output file
-p, --paths <in.ctp> Assembly file to load (can specify multiple times)
Type a command with no arguments to see usage. The following may also be useful:
Live chat (email me to fix a time):
Issues can be submitted on github. Pull requests welcome. Please add your name to the AUTHORS file. Code should compile on mac/linux with clang/gcc without errors or warnings.
More on the wiki
Unit tests are run with make test
and integration tests with cd tests; ./run
. Both of these test suites are run automatically with Travis CI when commits are pushed to GitHub.
Static analysis can be run with cppcheck:
cppcheck src
or with clang:
rm -rf bin/mccortex31
scan-build make RECOMPILE=1
Occasionally we also run Coverity Scan. This is done by pushing to the coverity_scan
branch on github, which triggers Travis CI to upload the latest code to Coverity.
git checkout coverity_scan
git merge develop
git checkout --ours .travis.yml
git checkout --ours configure
Bundled libraries may have different licenses:
Used in testing:
'Cortex with low memory and read threading' is currently unpublished. Please cite previous cortex_var papers: