cerebra 1.1.0

finds mutants in your scRNA-seq experiment

cerebra

What is cerebra?

This tool allows you to quickly extract meaningful variant information from a DNA or RNA sequencing experiment. If you're interested in learning what variants are present in your DNA/RNA samples, variant callers like GATK HaplotypeCaller can be used to generate variant calling format (VCF) files following a sequencing experiment. A VCF file looks like this:

##source=HaplotypeCaller
#CHROM POS ID REF ALT QUAL FILTER INFO FORMAT
chr1 631391 . C T 72.28 . AC=2;AF=1.00;AN=2;DP=2;ExcessHet=3.0103;FS=0.000;MLEAC=1;MLEAF=0.500;MQ=NaN;QD=25.36;SOR=2.303 GT:AD:DP:GQ:PL 1/1:0,2:2:6:84,6,0

Note that only a single VCF record is displayed here. A sequencing run can generate on the order of 10^8 unique VCF records, only a small portion of which contain meaningful biological signal. Thus drawing conclusions from VCF files remains a substantial challange. cerebra provides a fast and intuitive framework for summarizing VCF records across samples. It is comprised of three modules that do the following:

    1) remove germline variants from samples of interest        
    2) count the total number of variants in a given sample, on a per-gene basis           
    3) report peptide-level variants for each sample                 

cerebra gets its name from the eponymous X-men character, who had the ability to detect mutant individuals among the general public.

If you're working with tumor data, it might be a good idea to limit the variant search space to only known cancer variants. Therefore cerebra implements an optional method for restricting to variants also found in the COSMIC database.

This tool was developed for, but is certainly not limited to, single-cell RNA sequencing data.

• Free software: MIT license

What makes cerebra different from traditional VCF parsers?

Python libraries exist (i.e. PyVCF and vcfpy) for extracting information from VCF files, and GATK has its own tool for the task. In fact we integrate vcfpy into our tool. What makes cerebra different is that it reports the RNA transcript and amino acid change associated with each variant. GATK VariantsToTable produces a file that looks like:

CHROM    POS ID      QUAL    AC
 1        10  .       50      1
 1        20  rs10    99      10

Such a table contains only genomic (i.e. DNA-level) coordinates. Often the next question is what specific gene and peptide-level variants is each variant associated with? cerebra queries a reference genome (.fa) and annotation (.gtf) to match each DNA-level variant with its associated gene, probable transcript and probable peptide-level level variants. cerebra produces the following outfile:

$ python
> import json
> f = open(/pth/to/cerebra/output.json)
> data = json.load(f)
> print(json.dumps(data, indent=4))

{
    "CCN1": {
        "A16_B000563": [],
        "A1_B001546": [],
        "A1_B002531": [
            "ENSP00000398736.2:p.(Glu189=)"
        ],
        "A1_B002570": [],
        "A2_B002558": [],
        "A3_B000561": [
            "ENSP00000398736.2:p.(Arg209Trp)",
            "ENSP00000398736.2:p.(Ile90Val)"
        ],
        "A3_B000568": [],
        "A3_B001544": [
            "ENSP00000398736.2:p.(Ala82Thr)"
        ],
        "A3_B002090": [],
        "A3_B002531": [
            "ENSP00000398736.2:p.(Pro217Ser)"
        ]
    },
    "GOLGB1": {
        "A16_B000563": [],
        "A1_B001546": [
            "ENSP00000484083.1:p.?",
            "ENSP00000377275.3:p.(Ala1826Val)",
            "ENSP00000484083.1:p.(Gly1690Asp)",
            "ENSP00000484083.1:p.(Ala1746Val)",
            "ENSP00000377275.3:p.(Gly1770Asp)",
            "ENSP00000341848.5:p.(Thr911Ser)",
            "ENSP00000417767.1:p.(Thr782Ser)",
        ],
        "A1_B002531": [],
        "A1_B002570": [],
        "A2_B002558": [],
        "A3_B000561": [],
        "A3_B000568": [],
        "A3_B001544": [
            "ENSP00000377275.3:p.?",
            "ENSP00000341848.5:p.?",
            "ENSP00000484083.1:p.?"
        ],
        "A3_B002090": [],
        "A3_B002531": []
    }
}

Here CCN1 and GOLGB1 are gene names while A16_B000563, A1_B001546, A1_B002531,... are RNA-seq sample IDs. cerebra reports variants to every gene in the genome, for every sample in a given experiment. The ENSP**** numbers refer to Ensembl translation IDs -- unique identifiers that correspond to exactly one polypeptide in the Ensembl database. The strings enclosed in parentheses refer to the amino-acid level variant reported in that particular sample. For example the string Arg209Trp indicates that position 209 of this particular polypeptide should contain an Arg, but the experimental sample instead contains a Trp. cerebra adheres to HGVS sequence variant nomenclature in reporting amino-acid variants.

Features

germline-filter

If the research project is centered around a "tumor/pathogenic vs control" question, then germline-filter is the proper starting point. This module removes germline variants that are common between the control and the experimental tissue so as to not bias the results by including non-pathogenic variants. The user provides a very simple metadata file that indicates which experimental samples correspond to which control samples. For example:

experimental_sample_id,germline_sample_id
sample1,gl_sample1
sample2,gl_sample1
sample3,gl_sample2
sample4,gl_sample2
sample5,gl_sample2

There is also the option to limit the reported variants to those found in NCBI's dbSNP and the Wellcome Sanger Institute's COSMIC databases. This option is designed to give the user a higher degree of confidence in the pathogenic nature of each variant -- if independent experiments have reported a given variant in human tissue, there is a higher likilihood that it is pathogenic. The output of germline-filter is a set of trimmed-down VCF files.

If you have access to "control" tissue and your experimental question is concerned with differences between tumor/pathogenic tissue and control tissue, then germline-filter is the right place to start. germline-filter will produce a new set of VCFs, which you'll use for the next two steps. If you do not have access to "control" tissue, then proceed directly to count-variants or find-peptide-variants.

count-variants

The count-variants module reports the raw variant counts for every gene across every sample. The output is a CSV file that contains counts for each sample versus every gene in the genome.

find-peptide-variants

The find-peptide-variants module reports the peptide-level consequence of variants in the genome. If working with cancer samples, the user has the option to filter out all variants that are not found in the COSMIC database and are therefore unlikely to be pathogenic. VCF records are converted to peptide-level variants, and then ENSEMBL protein IDs, in acordance to the HGVS sequence variant nomenclature. The output is a heirarchically ordered text file (CSV or JSON) that reports the the Ensemble protein ID and the gene associated with each variant, for each experimental sample.

Variant callers are known to produce a great deal of false positives; the --report-coverage option is designed to give the user a greater degree of confidence in individual variant calls. If indicated this option will report raw counts for variant and wildtype reads at each variant loci. We reason that variants with a high degree of read support are less likely to be false positives.

We should stress that find-peptide-variants does not definitively report peptide-level variants but rather the likely set of peptide variants. Definitively reporting protein variants requires knowledge of alternate splicing -- this represents an open problem in scRNA-seq. For example, if a read picks up a variant in exon 2 of geneA, we can report each of the potential spliceforms of geneA that contain exon 2, but we cannot infer which of those particular spliceforms are actually present in our sample. Thus we report all possible spliceforms; determining the spliceform landscape of an individual cell from scRNA-seq is outside the scope of this project.

Note that all three modules take advantage of multiprocessing. Thus cerebra should scale better to high-memory machines with more cores, though it has been designed to run on everyday hardware.

Installation

To install the latest version from PyPi you'll first need to install a few system-specific dependencies.

For OSX:

sudo pip install setuptools
brew update
brew install openssl
brew install zlib

For Debian/Ubuntu:

sudo apt-get install autoconf automake make gcc perl zlib1g-dev libbz2-dev liblzma-dev libcurl4-gnutls-dev libssl-dev

Following that, you can install directly from PyPi.
pip install cerebra

If you prefer working with virtual environments you can clone from github and install with pip.

git clone https://github.com/czbiohub/cerebra.git
cd cerebra
conda create -n cerebra python=3.7
conda activate cerebra
pip install -e . 

Usage

cerebra should now be installed as a commandline executable. $ cerebra should return help information

Usage: cerebra  <command>

  a tool for fast and accurate summarizing of variant calling format (VCF)
  files

Options:
  -h, --help  Show this message and exit.

Commands:
  germline-filter    filter out common SNPs/indels between control/germline samples and samples of interest
  count-variants    count total number of variants in each sample, and report on a per-gene basis
  find-peptide-variants  report peptide-level SNPs and indels in each sample, and associated coverage

Note that the -h command will display usage information for each of the three commands.

An example workflow might look like this:

Step 1:
cerebra germline-filter --processes 2 --control_path /path/to/control/vcfs --experimental_path /path/to/experimental/vcfs --metadata /path/to/metadata/file --outdir /path/to/filtered/vcfs

Step 2:
cerebra count-variants --processes 2 --cosmicdb /optional/path/to/cosmic/database --refgenome /path/to/genome/annotation --outfile /path/to/output/file /path/to/filtered/vcfs/*

Step 3:
cerebra find-peptide-variants --processes 2 --cosmicdb /optional/path/to/cosmic/database --annotation /path/to/genome/annotation --genomefa /path/to/genome/fasta --report_coverage 1 --output /path/to/output/file /path/to/filtered/vcfs/*

Authors

This work was produced by Lincoln Harris, Rohan Vanheusden, Olga Botvinnik and Spyros Darmanis of the Chan Zuckerberg Biohub. For questions please contact [e-mail]

Contributing

We welcome any bug reports, feature requests or other contributions. Please submit a well documented report on our issue tracker. For substantial changes please fork this repo and submit a pull request for review.

Feel free to clone but NOTE this project is still a work in progress. You can find official releases on PyPi.