Assuming the input data are a list of VCF files (for examples, multiple-sample VCF files split by chromosomes or genes), this workflow runs annovar, intervar and snpEff to predict P/LP variants in two ways: 1. Clinvar+Intervar (CI); 2. Jung's criteria as descried in this paper (as I learned from Jung Kim ). The first method is straightforward, that is, the variant is predicted by as P/LP by either InterVar or ClinVar (with the CLINSTAT tag as "criteria_provided,_multiple_submitters,_no_conflicts", "practice_guideline", or "reviewed_by_expert_panel"). The second method is described in the paper. I named it "Jung's criteria" as I learned it from Dr. Jung Kim (and I have no idea how to call it ).
Assuming the input data are a list of VCF files (for examples, multiple-sample VCF files split by chromosomes or genes), this workflow runs annovar, intervar and snpEff to predict P/LP variants in two ways: 1. Clinvar+Intervar (CI); 2. Jung's criteria as descried in this paper. The first method is straightforward, that is, the variant is predicted by as P/LP by either InterVar or ClinVar (with the CLINSTAT tag as "criteria_provided,_multiple_submitters,_no_conflicts", "practice_guideline", or "reviewed_by_expert_panel"). The second method is a little complicated, as described in the paper. I named it "Jung's criteria" as I learned it from Dr. Jung Kim (and I have no idea how to call it in a better way :) ). Briefly, it is a union set of Clinvar+InterVar and two other variant sets based on the variant annotations from InterVar, AnnoVar and snpEff:
- variants with High impact (snpEff);
- variants meeting all the following criteria;
- Variants are predicted as VUS (variants of uncertain significance) by InterVar;
- Variants meets "3/4", "2/3", or "3/7"
- Predicted as ”damaging” missense mutations by 3 out of the 4 predictions;
- Predicted as "damaging" splice site mutations by 2 out of 3 predictions;
- Predicted as "damaging" missense or splice site mutations by 3 out of 7 predictions;
The overall process is as described in the diagram below:
Users may have a look of the R script Call_patho.R to understand how P/LP variants are actually predicted.
There are several steps taking in this Snakemake workflow as outlined in this diagram:
- Merge input vcf files into one file;
- 🔖 Only the first 8 columns of the VCF files are used so as to reduce the file size and facilitate the subsequent data processing.
- Split vcf files into multiple parts to speed up the annotation process.
- Annotate each split part by InterVar, Annovar and snpEff.
- Call P/LP variants using the R script Call_patho.R.
- There are two output files generated in this step: {chunk}.plp.txt and {chunk}.plp_slim.txt
- All the essential output from the variant annotators are kept in the plp.txt file (see this example).
- The plp_slim.txt file has only 8 columns (see this example):
- There are two output files generated in this step: {chunk}.plp.txt and {chunk}.plp_slim.txt
| vid | Gene.refGene | popmax_freq | PLP.clinvar | PLP.intervar | PLP.impactHigh | PLP.genomel | PLP.m_sp | PLP.jung |
|---|---|---|---|---|---|---|---|---|
| 4:186317998:C:A | ANKRD37;LRP2BP | 0 | FALSE | FALSE | FALSE | FALSE | FALSE | FALSE |
| 4:186318030:C:T | ANKRD37;LRP2BP | 0 | FALSE | FALSE | FALSE | FALSE | FALSE | FALSE |
| 4:186318071:G:A | ANKRD37;LRP2BP | 1.00E-04 | FALSE | FALSE | FALSE | FALSE | FALSE | FALSE |
| 4:186318186:G:A | LRP2BP | 0 | FALSE | FALSE | FALSE | FALSE | FALSE | FALSE |
| 4:186318261:G:A | LRP2BP | 8.00E-04 | FALSE | FALSE | FALSE | FALSE | FALSE | FALSE |
| 4:186318265:A:G | LRP2BP | 6.84E-05 | FALSE | FALSE | FALSE | FALSE | FALSE | FALSE |
| 4:186318282:T:TG | LRP2BP | 0 | FALSE | FALSE | FALSE | FALSE | FALSE | FALSE |
- Below is the description of the columns in the slim file
| Column ID | Description |
|---|---|
| vid | Variant ID in the format of CHROM:POS:REF:ALT |
| Gene.refGene | Gene ID |
| popmax_freq | Maximum AF among the subpopulations (derived from "AF_popmax" and "non_cancer_AF_popmax") |
| PLP.clinvar | P/LP defined by ClinVar |
| PLP.intervar | P/LP defined by InterVar |
| PLP.impactHigh | P/LP defined by high impact |
| PLP.genomel | P/LP defined (impact == 'HIGH') | (impact == "MODERATE" & Polyphen2_HDIV_pred == 'D') |
| PLP.m_sp | P/LP defined as Damaging missense and/or splite site mutations |
| PLP.jung | P/LP defined by PLP.clinvar | PLP.intervar | PLP.impactHigh | (intervar=="Uncertain significance" & PLP.m_sp) |
Even though Snakemake module is available at Biowulf, a conda installed Snakemake is recommend. In this study, Snakemake (Version 7.3.7) was used.
Conda is also used in the workflow to create a working environment to call P/LP using R.
- workflow/envs/plp.yaml
channels:
- conda-forge
dependencies:
- r-base =4.2.1
- r-bigreadr =0.2.4
- r-rlang =1.0.6
- r-tidyverse =1.3.2This workflow has taken full advantage of the modules installed at Biowulf (non-biowulf users may use conda instead):
- bcftools/1.13
- annovar/2020-06-08
- snpEff/5.1d
- csvkit/1.0.7
By the time of development, the latest module of intervar at Biowulf is Version 2.1.3. So, I had installed InterVar (Version 2.2.1) locally, with the InterVar database under /home/zhuw10/git/InterVar-2.2.1/.
/home/zhuw10/git/InterVar-2.2.1/Intervar.py -i {input} -b {params.
genome} -d $ANNOVAR_DATA/{params.genome} -o {params.prefix} -t /home/zhuw10/git/InterVar-2.2.1/intervardb --table_annovar=$ANNOVAR_HOME/table_annova
r.pl --convert2annovar=$ANNOVAR_HOME/convert2annovar.pl --annotate_variation=$ANNOVAR_HOME/annotate_variation.pl 2>{log}Currently, the module "intervar/2.2.1" is available at Biowulf, so users may just update: /home/zhuw10/git/InterVar-2.2.1/Intervar.py and /home/zhuw10/git/InterVar-2.2.1/intervardb accordingly:
module show intervar/2.2.1
----------------------------------------------------------------------
/usr/local/lmod/modulefiles/intervar/2.2.1.lua:
----------------------------------------------------------------------
help([[This module sets up the environment for using InterVar.
]])
whatis("InterVar: Clinical Interpretation of Genetic Variants")
whatis("Version: 2.2.1")
setenv("INTERVAR_HOME","/usr/local/apps/intervar")
setenv("INTERVAR_BIN","/usr/local/apps/intervar/2.2.1/bin")
setenv("INTERVAR_TEST","/usr/local/apps/intervar/TEST_DATA/example")
setenv("INTERVAR_DATA","/usr/local/apps/intervar/DOWNLOADS/InterVar")
prepend_path("PATH","/usr/local/apps/intervar/2.2.1/bin")
prepend_path("PATH","/usr/local/apps/annovar/2020-06-08")
setenv("ANNOVAR_HOME","/usr/local/apps/annovar/2020-06-08")
setenv("ANNOVAR_DATA","/fdb/annovar/2020-06-08")
setenv("ANNOVAR_DATA_CURRENT","/fdb/annovar/current")All input files are configured in one yaml file for the workflow:
- Example of the configure file of hg38
# given a list vcf file under the folder
vcf_input_dir: "/data/zhuw10/ukbb/dnanexus/pvcf2"
### output_dir is sufficient to keep data unique and output_prefix is to for the merged file
output_dir: "/data/zhuw10/ukbb/dnanexus/batch2"
output_prefix: "ukb_batch2"
ref: "/data/zhuw10/ukbb/ref/GRCh38_full_analysis_set_plus_decoy_hla.fa"
genome: "hg38"
split_total: 100
# for GRCh38.99
snpEff_db: "GRCh38.99" - Example of the configure file of hg19
# given a list vcf file under the folder
vcf_input_dir: "/data/zhuw10/ukbb/data/genomel_vcf"
### output_dir is sufficient to keep data unique and output_prefix is to for the merged file
output_dir: "/data/zhuw10/ukbb/dnanexus/genomel_out"
output_prefix: "prj1"
ref: "/data/zhuw10/ukbb/ref/Homo_sapiens_assembly19.fasta"
genome: "hg19"
split_total: 100
# for GRCh38.99
snpEff_db: "GRCh37.87" Overall, there are several input files are required:
- VCF input files under the specified folder vcf_input_dir.
- The reference genome (specified by ref) in the fasta format with fai index.
Besides, there are several parameters defined in the configure file.
- output_dir, specified the output folder.
- genome, specified the resource bundle used by annovar and intervar.
- output_prefix, the vcf filed will be processed and merged as one file output_dir/{output_prefix}.vcf.gz.
- snpEff_db, the snpEff database used in the snpEff annotation. It may need to be updated if new database is available.
- split_total, specified the number of parts to be split and processed in parallel.
Assuming you have git clone this repo under $PLP_prediction_workflow, and provided input files and the configure file properly, it is simple to launch the workflow at Biowulf using the wrapper script run_it2.sh:
cd $PLP_prediction_workflow/workflow
sbatch -J gnomad --export=ALL --mem=12g -p norm -o ${PWD}/slurm-%j.out -e ${PWD}/slurm-%j.err --time=24:00:00 --wrap='./run_it2.sh ../config/gnomad_config.yaml'The running time is an important factor to be considered in any NGS tools. Given split_total like 100, gnomAD annotation can be accomplished in hours.
This workflow is dedicated to the GenoMEL project. Some steps might not be needed for most uses, for example, like low, proxy and syn. I kept them here as examples about how to use csvkit to filter the candidate variants from plp.txt files in the way your may prefer:
### Identify synonymous variants under the AF specified by {params.min_freq}
csvsql --query "select vid, [Gene.refGene] from '`basename {input} .txt`' where (popmax_freq IS NULL or popmax_freq < {params.min_freq} ) AND [PLP.jung] = 0 AND ([ExonicFunc.refGene] = 'synonymous SNV' ) " {input} > {output}
### Identify P/LP as Jung suggested under the AF specified by {params.min_freq}
csvsql --query "select vid, [Gene.refGene] from '`basename {input} .txt`' where (popmax_freq IS NULL or popmax_freq < {params.min_freq} ) AND ([PLP.jung]= 1) " {input} > {output}