A Nextflow Genome-Wide Association Study (GWAS) Pipeline
$ git clone https://github.com/montilab/nf-gwas-pipelineWe have provided multiple toy datasets for testing the pipeline and ensuring all paths and dependencies are properly setup. To set the toy data paths to your local directory, run the following script.
$ cd nf-gwas-pipeline
$ python utils/paths.py Nextflow requires a POSIX compatible system (Linux, OS X, etc.) and Java 8 (or later, up to 11) to be installed. Once downloaded, optionally make the nextflow file accessible by your $PATH variable so you do not have to specify the full path to nextflow each time.
$ curl -s https://get.nextflow.io | bash
We have created a pre-built Docker image with all of the dependencies installed. To get started, first make sure Docker is installed. Then pull down the image onto your local machine.
$ docker pull montilab/gwas:latest
Optionally you could build this image yourself from the Dockerfile which specifies all of the dependencies required. Note: This might take a while!
$ docker build --tag montilab/gwas:latest .
$ ./nextflow gwas.nf -c gwas.config -with-docker montilab/gwas
N E X T F L O W ~ version 19.04.1
Launching `gwas.nf` [jolly_fermi] - revision: 46311ebd05
-
G W A S ~ P I P E L I N E
================================
indir : <YOUR PATH>/data/
outdir : <YOUR PATH>/results
vcf : <YOUR PATH>/data/toy_vcf.csv
pheno : <YOUR PATH>/data/pheno_file_logistic.csv
snpset : <YOUR PATH>/data/snpset.txt
phenotype : outcome
covars : age,sex,PC1,PC2,PC3,PC4
model : logistic
test : Score
ref : hg19
-
[warm up] executor > local
executor > local (141)
[60/b5b95e] process > qc_miss [100%] 22 of 22 âś”
[11/fa0fbd] process > annovar_ref [100%] 1 of 1 âś”
[8f/25f8fa] process > qc_mono [100%] 22 of 22 âś”
[82/069a6d] process > vcf_to_gds [100%] 22 of 22 âś”
[3e/819e86] process > merge_gds [100%] 1 of 1 âś”
[c3/f23390] process > nullmod_skip_pca_grm [100%] 1 of 1 âś”
[ed/91344b] process > gwas_skip_pca_grm [100%] 22 of 22 âś”
[b4/3aea3e] process > caf_by_group_skip_pca_grm [100%] 22 of 22 âś”
[e2/3c778d] process > merge_by_chr [100%] 22 of 22 âś”
[fe/33ebd4] process > combine_results [100%] 1 of 1 âś”
[8b/2020d3] process > annovar_input [100%] 1 of 1 âś”
[61/3a373f] process > plot [100%] 1 of 1 âś”
[66/6f4246] process > annovar [100%] 1 of 1 âś”
[85/d4266b] process > add_annovar [100%] 1 of 1 âś”
[9e/4fc2fe] process > report [100%] 1 of 1 âś”
Completed at: 15-Oct-2020 17:30:28
Duration : 44.1s
CPU hours : 0.1
Succeeded : 141If you are running the pipeline on a HPC that does not support docker (BU’s Shared Computing Cluster), you can load the dependencies and run the pipeline as follows. (In addition, you need to install following R packages: SeqArray, GENESIS, Biobase, SeqVarTools, dplyr, SNPRelate, ggplot2, data.table, reshape2, latex2exp, knitr, EBImage, GenomicRanges, TxDb.Hsapiens.UCSC.hg19.knownGene, GMMAT, ezknitr)
$ module load R/4.1.1
$ module load vcftools/0.1.16
$ module load bcftools/1.10.2
$ module load plink/2.00a1LM
$ module load annovar/2018apr
$ module load pandoc/2.5
nextflow gwas.nf -c gwas.config
- The first column should be the unique ID for subjects
- Names of the columns and numbers of columns are not fixed
- The group variable is optional but should be a categorical variable if called
- Longitudinal phenotype file shoud be in long-format
- If the pca_grm process is turned-off, PCs should present in the phenotype file to be called
example: ./data/pheno_file_linear.csv
./data/pheno_file_logistic.csv
./data/1KG_pheno_linear.csv
./data/1KG_pheno_logistic.csv
./data/1KG_pheno_longitudinal.csv
pheno.dat <- read.csv("data/pheno_file_linear.csv")
kable(head(pheno.dat))| ID | outcome | age | sex | PC1 | PC2 | PC3 | PC4 | group |
|---|---|---|---|---|---|---|---|---|
| 202578640192_R09C01_202578640192_R09C01 | -1.1259198 | 53.03908 | F | -0.0048 | 0.0211 | 0.0389 | -0.0168 | group2 |
| 202579010063_R05C02_202579010063_R05C02 | -2.3237168 | 59.39922 | F | -0.0383 | -0.0157 | 0.0061 | 0.0108 | group2 |
| 202578650131_R04C02_202578650131_R04C02 | 0.0589976 | 22.27178 | F | -0.0356 | -0.0149 | -0.0159 | 0.0113 | group3 |
| 202582730083_R09C01_202582730083_R09C01 | 0.9995060 | 68.75518 | M | 0.0079 | -0.0043 | -0.0103 | -0.0257 | group1 |
| 202578640258_R03C02_202578640258_R03C02 | -0.9547252 | 23.48552 | F | 0.0148 | -0.0079 | 0.0120 | 0.0058 | group3 |
| 202578650131_R05C01_202578650131_R05C01 | 0.5786668 | 11.09063 | M | 0.0065 | 0.0030 | -0.0128 | 0.0157 | group3 |
- vcf.gz files at least contains the GT column
- The ID column would end up being the snpID in the final output
- vcf.file should contain DS column to use dosages in GWAS (imputed=T)
example: ./data/vcf/vcf_file1.vcf.gz
./data/1KG_vcf/1KG_phase3_subset_chr1.vcf.gz
- Two-column csv file mapping the prefix to the vcf.gz files
- The results for each chromosome will be names be the corresponding prefix
- NO header
example: ./data/toy_vcf.csv
./data/1KG_vcf.csv
map.dat <- read.csv("./data/toy_vcf.csv", header=F)
kable(head(map.dat))| V1 | V2 |
|---|---|
| chr_1 | /nf-gwas-pipeline/data/vcf/vcf_file1.vcf.gz |
| chr_2 | /nf-gwas-pipeline/data/vcf/vcf_file2.vcf.gz |
| chr_3 | /nf-gwas-pipeline/data/vcf/vcf_file3.vcf.gz |
| chr_4 | /nf-gwas-pipeline/data/vcf/vcf_file4.vcf.gz |
| chr_5 | /nf-gwas-pipeline/data/vcf/vcf_file5.vcf.gz |
| chr_6 | /nf-gwas-pipeline/data/vcf/vcf_file6.vcf.gz |
- Two column txt file seperated by “,”
- First column shoud be chromosome and second column be physical position with fixed header “chr,pos”
example: ./data/snpset.txt
snp.dat <- fread("./data/snpset.txt")
kable(head(snp.dat))| chr | pos |
|---|---|
| 1 | 1165522 |
| 1 | 1176433 |
| 1 | 1179532 |
| 1 | 1188944 |
| 1 | 1781220 |
| 2 | 1018108 |
- A symmetric matrix saved in rds format with both columns being subjects
- Can be replaced by 2*kinship matrix
grm <- readRDS("./data/grm.rds")
kable(grm[1:5,1:5])| HG00110 | HG00116 | HG00120 | HG00128 | HG00136 | |
|---|---|---|---|---|---|
| HG00110 | 1.0332116 | -0.0179534 | 0.0070812 | -0.0114037 | -0.0122968 |
| HG00116 | -0.0179534 | 0.9901158 | 0.1161200 | -0.0369330 | -0.0204240 |
| HG00120 | 0.0070812 | 0.1161200 | 0.9772376 | -0.0595185 | -0.0337373 |
| HG00128 | -0.0114037 | -0.0369330 | -0.0595185 | 0.9500809 | -0.0373967 |
| HG00136 | -0.0122968 | -0.0204240 | -0.0337373 | -0.0373967 | 0.9740444 |
pheno.dat <- read.csv("./data/1KG_pheno_logistic.csv")
kable(head(pheno.dat))| sample.id | Population | sex | outcome |
|---|---|---|---|
| HG00110 | GBR | F | 1 |
| HG00116 | GBR | M | 1 |
| HG00120 | GBR | F | 0 |
| HG00128 | GBR | F | 1 |
| HG00136 | GBR | M | 0 |
| HG00137 | GBR | F | 0 |
map.dat <- read.csv("./data/1KG_vcf.csv", header=F)
kable(head(map.dat))| V1 | V2 |
|---|---|
| chr_1 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr1.vcf.gz |
| chr_2 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr2.vcf.gz |
| chr_3 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr3.vcf.gz |
| chr_4 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr4.vcf.gz |
| chr_5 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr5.vcf.gz |
| chr_6 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr6.vcf.gz |
See mapping file
run with .config file:
nextflow run gwas.nf -c $PWD/configs/gwas_1KG_logistic.config
run with equivalent command:
nextflow run gwas.nf --vcf_list $PWD/data/1KG_vcf.csv --pheno $PWD/data/1KG_pheno_logistic.csv --phenotype outcome --covars sex,PC1,PC2,PC3,PC4 --pca_grm --model logistic --test Score --gwas --group Population --min_maf 0.1 --max_pval_manhattan 0.5 --max_pval 0.05 --ref_genome hg19pheno.dat <- read.csv("./data/1KG_pheno_linear.csv")
kable(head(pheno.dat))| sample.id | Population | sex | outcome |
|---|---|---|---|
| HG00110 | GBR | F | 1.2114051 |
| HG00116 | GBR | M | 1.4196076 |
| HG00120 | GBR | F | 0.0119097 |
| HG00128 | GBR | F | 0.6800792 |
| HG00136 | GBR | M | -2.3179815 |
| HG00137 | GBR | F | -1.4958842 |
map.dat <- read.csv("./data/1KG_vcf.csv", header=F)
kable(head(map.dat))| V1 | V2 |
|---|---|
| chr_1 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr1.vcf.gz |
| chr_2 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr2.vcf.gz |
| chr_3 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr3.vcf.gz |
| chr_4 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr4.vcf.gz |
| chr_5 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr5.vcf.gz |
| chr_6 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr6.vcf.gz |
See mapping file
run with .config file:
nextflow run gwas.nf -c $PWD/configs/gene_1KG_linear.config
run with equivalent command:
nextflow run gwas.nf --vcf_list $PWD/data/1KG_vcf.csv --pheno $PWD/data/1KG_pheno_linear.csv --phenotype outcome --covars PC1,PC2,PC3,PC4 --pca_grm --model linear --test Score --gene_based --group Population --max_pval 0.01 --ref_genome hg19pheno.dat <- read.csv("./data/1KG_pheno_longitudinal.csv")
kable(head(pheno.dat))| sample.id | Population | sex | age | delta.age | outcome |
|---|---|---|---|---|---|
| HG00110 | GBR | F | 46 | 0 | 1.2114051 |
| HG00110 | GBR | F | 53 | 7 | 3.1471562 |
| HG00116 | GBR | M | 51 | 0 | 1.4196076 |
| HG00116 | GBR | M | 57 | 6 | 1.9318303 |
| HG00120 | GBR | F | 49 | 0 | 0.0119097 |
| HG00120 | GBR | F | 57 | 8 | 3.1782473 |
map.dat <- read.csv("./data/1KG_vcf.csv", header=F)
kable(head(map.dat))| V1 | V2 |
|---|---|
| chr_1 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr1.vcf.gz |
| chr_2 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr2.vcf.gz |
| chr_3 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr3.vcf.gz |
| chr_4 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr4.vcf.gz |
| chr_5 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr5.vcf.gz |
| chr_6 | /nf-gwas-pipeline/data/1KG_vcf/1KG_phase3_subset_chr6.vcf.gz |
See mapping file
run with .config file:
nextflow run gwas.nf -c $PWD/configs/gwla_1KG_linear_slope.config
run with equivalent command:
nextflow run gwas.nf --vcf_list $PWD/data/1KG_vcf.csv --pheno $PWD/data/1KG_pheno_longitudinal.csv --phenotype outcome --covars sex,age,PC1,PC2,PC3,PC4 --pca_grm --model linear --test Score --longitudinal --random_slope delta.age --group Population --min_maf 0.1 --max_pval_manhattan 0.5 --max_pval 0.01 --ref_genome hg19you can see explanations for all parameters with the help command:
nextflow gwas.nf --help
N E X T F L O W ~ version 19.04.1
Launching `gwas.nf` [tiny_venter] - revision: c9ded642f7
USAGE:
Mandatory arguments:
--vcf_list String Path to the two-column mapping csv file: id , file_path
--pheno String Path to the phenotype file
--phenotype String Name of the phenotype column
Optional arguments:
--gds_input Logical If true, ignore vcf input, start with GDS files and skip qc_miss, qc_mono, vcf_to_gds steps
--gds_list String Path to the two-column mapping gds file: id , file_path
--outdir String Path to the master folder to store all results
--covars String Name of the covariates to include in analysis model separated by comma (e.g. "age,sex,educ")
--qc Logical If true, run qc_miss(filter genotypes called below max_missing) and qc_mono (drop monomorphic SNPs)
--max_missing Numeric Threshold for qc_miss (filter genotypes called below this value)
--pca_grm Logical If true, run PCAiR (generate PCA in Related individuals) and PCRelate (generate genomic relationship matrix)
--snpset String Path to the two column txt file separated by comma: chr,pos (can only be effective when pca_grm = true)
--grm String Path to the genomic relationship matrix (can only be effective when pca_grm = false)
--model String Name of regression model for gwas: "linear" or "logistic"
--test String Name of statistical test for significance: "Score", "Score.SPA", "BinomiRare" and "CMP" (details see https://rdrr.io/bioc/GENESIS/man/assocTestSingle.html)
--gwas Logical If true, run gwas
--imputed Logical If true, use dosages in regression model (DS columns needed in input vcf files)
--gene_based Logical If true, run aggregate test for genes based on hg19 reference genome
--max_maf Numeric Threshold for maximun minor allele frequencies of SNPs to be aggregated
--method String Name of aggregation test method: "Burden", "SKAT", "fastSKAT", "SMMAT" or "SKATO"
--longitudinal Logical If true, run genome-wide longitudianl analysis
--random_slope String if set to "null", random intercept only model is run; else run random slope and random intercept model
--group String Name of the group variable based on which the allele frequencies in each subgroup is calculated (can be left empty)
--dosage Logical If true, also calculate dosages in addition to allele frequencies (can be very slow with large single gds input)
--min_maf Numeric Threshold for minimun minor allele frequencies of SNPs to include in QQ- and Manhattan-plot
--max_pval_manhattan Numeric Threshold for maximun p-value of SNPs to show in Manhattan-plot
--max_pval Numeric Threshold for maxumun p-value of SNPs to annotate
--ref_genome String Name of the reference genome for annotation: hg19 or hg38