CADD-SV

CADD-SV – a framework to score the effect of structural variants

Here, we describe CADD-SV, a method to retrieve a wide set of annotations in the range and vicinity of a SV. Our tool computes summary statistics and uses a trained machine learning model to differentiate deleterious from neutral variants. In training, we use human and chimpanzee derived alleles as proxy-neutral and contrast them with matched simulated variants as proxy-pathogenic. This approach has proven powerful in the interpretation of SNVs (CADD, https://cadd.gs.washington.edu). We show that CADD-SV scores correlate with known pathogenic variants in individual genomes and allelic diversity.

Pre-requirements

Conda

The pipeline depends on Snakemake, a workflow management system that wraps up all scripts and runs them highly automated, in various environments (workstations, clusters, grid, or cloud). Further, we use Conda as software/dependency management tool. Conda can install snakemake and all neccessary software with its dependencies automatically. Conda installation guidelines can be found here:

https://conda.io/projects/conda/en/latest/user-guide/install/index.html

Snakemake

After installing Conda, you install Snakemake using Conda and the environment.yaml provided in this repository. For this purpose, please clone or download and uncompress the repository first. Then change into the root folder of the local repository.

git clone https://github.com/kircherlab/CADD-SV
cd CADD-SV

We will now initiate the Conda environment, which we will need for getting the Snakemake workflow invoked. Using this environment (run.caddsv) snakemake will be installed

conda env create -n run.caddsv --file environment.yaml

The second conda environment (envs/SV.yml), containing all packages and tools to run CADD-SV, will be installed automatically during the first run. This can take some time.

Annotations

CADD-SV depends on various annotations to provide the model with its necessary input features. CADD-SV automatically retrieves and transforms these annotations (see Snakefile) and combines them in bed-format at /desired-sv-set/matrix.bed

Annotations can be downloaded and expanded individually. However, to run CADD-SV with the pre-trained model and to minimize runtime and memory failures use the annotation sets as stored at https://kircherlab.bihealth.org/download/CADD-SV/

wget https://kircherlab.bihealth.org/download/CADD-SV/v1.1/dependencies.tar.gz
tar -xf dependencies.tar.gz

Config

Almost ready to go. After you prepared the files above, you may need to adjust the name of your dataset in the config.yml.

List of required input files

Models and scripts as cloned from this GIT repository
Annotations in the annotations/ folder
CADD-SV scores SV in a coordinate sorted BED format on the GRCh38 genome build. The type of SV needs to be included for each variant in the 4th column. We recommend to split files containing more than 10,000 SVs into smaller files. An example input file can be found in input/. The file needs to have the suffix id_. If you plan to process variants from another genome build or SVs in VCF format, see below.

Running the pipeline

Ready to go! If you run the pipeline on a cluster see the cluster.json for an estimate of minimum resource requirements for the individual jobs. Note that this depends on your dataset size so you may have to adjust this.

To start the pipeline:

conda activate run.caddsv
# dry run to see if everything works
snakemake  --use-conda --configfile config.yml -j 4 -n
# run the pipeline
snakemake  --use-conda --configfile config.yml -j 4

Output files

The pipeline outputs your SV set containing all annotations in BED format in a folder named output containing the CADD-SV and two raw scores in rows 6-8. Further information about individual annotations are kept in a subfolder named after your input dataset.

Further Information

Annotations

CADD-SV integrates different annotations, here some links to its annotation sources. A complete list can be found as Suppl. Table 1 of the manuscript/pre-print.

Integrated Scores

CADD (https://krishna.gs.washington.edu/download/CADD/bigWig/)
LINSIGHT (http://compgen.cshl.edu/LINSIGHT/LINSIGHT.bw)

Species conservation and constraint metrics

PhastCons (http://hgdownload.cse.ucsc.edu/goldenpath/hg38/)
Syntenic regions (http://webclu.bio.wzw.tum.de/cgi-bin/syntenymapper/get-species-list.py)
GERP score (http://mendel.stanford.edu/SidowLab/downloads/gerp/)

Population and disease constraint metrics

pLI score (ftp://ftp.broadinstitute.org/pub/ExAC_release/release1/manuscript_data/forweb_cleaned_exac_r03_march16_z_data_pLI.txt.gz)
Conserved coding regions (CCR) (https://www.nature.com/articles/s41588-018-0294-6?WT.feed_name=subjects_population-genetics)
DDD Happloinsufficiency (https://decipher.sanger.ac.uk/files/downloads/HI_Predictions_Version3.bed.gz)

Epigenetic and regulatory activity

Encode Features such as Histon Modifications and DNase and RNase-seq (https://www.encodeproject.org/help/batch-download/)
GC content (http://hgdownload.cse.ucsc.edu/gbdb/hg38/bbi/gc5BaseBw/gc5Base.bw)
ChromHMM states of ENCODE cell lines (http://compbio.mit.edu/ChromHMM/)

3D genome organization

CTCF (http://genome.cshlp.org/content/suppl/2012/08/28/22.9.1680.DC1/Table_S2_Location_of_ChIP-seq_binding_positions_in_19_cell_lines.txt)
Encode-HiC data (https://www.encodeproject.org/search/?type=Experiment&assay_term_name=HiC&replicates.library.biosample.donor.organism.scientific_name=Homo%20sapiens&status=released)
Enhancer-promoter-links from FOCS (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5930446/)
Frequently Interacting Regulatory Elements (FIREs) (https://www.sciencedirect.com/science/article/pii/S2211124716314814)
Directionality index for HiC data from various datasets (https://www.genomegitar.org/processed-data.html)
DeepC saliencies score (http://userweb.molbiol.ox.ac.uk/public/rschwess/container_for_ucsc/data/deepC/saliency_scores/saliencies_merged_gm12878_5kb.bw)

Gene and element annotations

Ensembl-gff3 genebuild 96 (ftp://ftp.ensembl.org/pub/release-96/gff3/homo_sapiens/Homo_sapiens.GRCh38.96.chr.gff3.gz)
Fantom5 enhancers (https://zenodo.org/record/556775#.Xkz3G0oo-70)

Converting VCF and other genome builds

If you want to score SVs in a VCF format or your SVs are not in GRCh38 genomebuild coordinates: We provide an environment to handle this. It uses the SURVIVOR tools (https://github.com/fritzsedlazeck/SURVIVOR).

conda env create -n prepBED --file envs/prepBED.yml

To convert your VCF into BED format run:

conda activate prepBED
SURVIVOR vcftobed input.vcf 0 -1 output.bed
cut -f1,2,6,11 output.bed > beds/set_id.bed

To lift hg19 coordinates to GRCh38 apply the following steps:

conda activate prepBED
liftOver beds/setname_hg19_id.bed dependencies/hg19ToHg38.over.chain.gz beds/setname_id.bed beds/setname_unlifted.bed