1 Introduction

gwasurvivr can be used to perform survival analyses of imputed genotypes from Sanger and Michigan imputation servers and IMPUTE2 software. This vignette is a tutorial on how to perform these analyses. This package can be run locally on a Linux, Mac OS X, Windows or conveniently batched on a high performing computing cluster. gwasurvivr iteratively processes the data in chunks and therefore intense memory requirements are not necessary.
gwasurvivr package comes with three main functions to perform survival analyses using Cox proportional hazard (Cox PH) models depending on the imputation method used to generate the genotype data:

  1. michiganCoxSurv: Performs survival analysis on imputed genetic data stored in compressed VCF files generated via Michigan imputation server.
  2. sangerCoxSurv: Performs survival analysis on imputed genetic data stored in compressed VCF files generated via Sanger imputation server.
  3. impute2CoxSurv: Performs survival analysis on imputed genetic data from IMPUTE2 output.
  4. plinkCoxSurv: Performs survival analysis on directly typed genetic data from PLINK files (.BED, .BIM, and .FAM)
  5. gdsCoxSurv: Performs survival analysis on genetic data that user has already converted from IMPUTE2 format to GDS format.

All functions fit a Cox PH model to each SNP including other user defined covariates and will save the results as a text file directly to the disk that contains survival analysis results. gwasurvivr functions can also test for interaction of SNPs with a given covariate. See examples for further details.

1.1 Main input arguments

All three functions require the following main arguments:

  • vcf.file: A character string giving the path to genotype data file (impute.file for IMPUTE2; b.file for PLINK)
  • covariate.file: A data frame comprising sample IDs (that link to the genotype data samples), phenotype (time, event) and additional covariate data
  • id.column: A character string providing sample ID column name in covariate.file
  • time.to.event: A character string that contains the time column name in covariate.file
  • event: Character string that contains the event column name in covariate.file
  • covariates: Character vector with contains column names in covariate.file of covariates to include in model
  • out.file: A character string giving output file name

Further arguments can be passed depending on the user preference. For example, user can define minor allele frequency (MAF) or info/R2 score threshold to filter out SNPs that have low MAF or info/R2 score to avoid false-positive signals and to reduce computing time. User can also define a subset of samples to be analyzed by providing the set of sample IDs. Users can also control how chunk size – the number of rows (SNPs) to include for each iteration.

IMPORTANT: In the covariate.file, categorical variables need to be converted to indicator (dummy) variables and be of class numeric. Ordinal variables represented as characters, ie “low”, “medium” and “high” should be converted to the appropriate numeric values as well.

1.2 Main output format

The output for the 3 main functions in gwasurvivr are very similar but with subtle differences. In general the output includes the following main fields: RSID, TYPED, CHR, POS, REF, ALT, Allele Frequencies*, INFO*, PVALUES, HRs, HR confidence intervals, coefficient estimates, standard errors, Z-statistic, N, and NEVENT. Allele frequencies and INFO differ by the input imputation software.

Note: Invoking the inter.term argument for any of the functions will make the PVALUE and HRs and HR confidence intervals represent the INTERACTION term and not the SNP alone.

The non-software specific fields are summarized below:

Column Description
RSID SNP ID
CHR Chromosome number
POS Genomic Position (BP)
REF Reference Allele
ALT Alternate Allele
SAMP_FREQ_ALT Alternate Allele frequency in sample being tested
SAMP_MAF Minor allele frequency in sample being tested
PVALUE P-value of single SNP or interaction term
HR Hazard Ratio (HR)
HR_lowerCI Lower bound 95% CI of HR
HR_upperCI Upper bound 95% CI of HR
COEF Estimated coefficient of SNP
SE.COEF Standard error of coefficient estimate
Z Z-statistic
N Number of individuals in sample being tested
NEVENT Number of events that occurred in sample being tested

The software specific fields are summarized below:
1. michiganCoxSurv unique output columns are AF, MAF, R2, ER2. They are summarized below.

Column Description
TYPED Imputation status: TRUE (SNP IS TYPED)/FALSE (SNP IS IMPUTED)
AF Minimac3 output Alternate Allele Frequency
MAF Minimac3 output of Minor Allele Frequency
R2 Imputation R2 score (minimac3 \(R^2\))
ER2 Minimac3 ouput empirical \(R^2\)

Please see Minimac3 Info File for details on output

  1. sangerCoxSurv
Column Description
TYPED Imputation status: TRUE (SNP IS TYPED)/FALSE (SNP IS IMPUTED)
RefPanelAF HRC Reference Panel Allele Frequency
INFO Imputation INFO score from PBWT
  1. impute2CoxSurv
Column Description
TYPED --- is imputed, repeated RSID is typed
A0 Allele coded 0 in IMPUTE2
A1 Allele coded 1 in IMPUTE2
exp_freq_A1 Expected allele frequency of alelle code A1

More statistics can be printed out by invoking the print.covs argument and setting it to print.covs=all (single SNP/SNP*covariate interaction) or print.covs=some (SNP*covariate ineraction). These options are available mainly for modeling purposes (covariate selection) and aren’t suggested for very large analyses as it will greatly increase the number of columns in the output, depending on how many covariates users are adjusting for.

2 Getting started

Install gwasurvivr from the Sucheston-Campbell Lab Github repository using devtools.

devtools::install_github("suchestoncampbelllab/gwasurvivr")

Or please install from the devel branch of Bioconductor (R version >= 3.5)

if (!requireNamespace("BiocManager", quietly=TRUE))
    install.packages("BiocManager")
BiocManager::install("gwasurvivr", version = "devel")    

2.1 Dependencies

Note: This package depends on GWASTools which uses netcdf framework on linux. Therefore, for linux users, please install libnetcdf-dev and netcdf-bin before installing gwasurvivr. These linux libraries may already installed on an academic computing cluster.

CRAN packages:
1. ncdf4
2. matrixStats
3. parallel
4. survival

install.packages(c("ncdf4", "matrixStats", "parallel", "survival"))

Bioconductor packages:
1. GWASTools
2. VariantAnnotation
3. SummarizedExperiment 4. SNPRelate

BiocManager::install("GWASTools")    
BiocManager::install("VariantAnnotation")    
BiocManager::install("SummarizedExperiment")    
BiocManager::install("SNPRelate")    

Load gwasurvivr.

library(gwasurvivr)

2.2 User settings: parallelization setup

gwasurvivr uses parallel package for its internal parallelization to fit the Cox PH models. Users are not required to define a parallelization setup, by default gwasurvivr functions will detect the user’s operating system and set the cluster object to FORK if the platform is Linux/OS X and to SOCK if Windows. However, parallelization settings can be modified by the user if needed. Users are given two ways to define their cluster settings:

1. Setting the number of cores to be used:

Linux/OS X users can run analyses on a prespecified number of cores by setting the option in the R session as shown below. This option should be defined in the R session before running any of the gwasurvivr functions. Here we decide to use 4 cores. This option is not available to Windows users.

options("gwasurvivr.cores"=4)

2. Providing a user defined cluster object

To modify more settings, users can also provide a “cluster object” to any of the gwasurvivr functions. The cluster object can be generated via makeCluster, makePSOCKcluster, makeForkCluster functions from parallel package or similar cluster object functions from snow or snowfall packages. This method can be applied on any operating system. User can create a cluster object before running any of the functions and pass the cluster object to the clusterObj argument as shown below. For further details see ??parallel::parallel.

library(parallel)
cl <- makeCluster(detectCores())

impute2CoxSurv(..., clusterObj=cl)
sangerCoxSurv(..., clusterObj=cl)
michiganCoxSurv(..., clusterObj=cl)

3 R Session Examples

3.1 Michigan Imputation Server

Michigan Imputation Server pre-phases typed genotypes using HAPI-UR, SHAPEIT, or EAGLE (default is EAGLE2), imputes using Minimac3 imputation engine and outputs Blocked GNU Zip Format VCF files (.vcf.gz). These .vcf.gz files are used as input for gwasurvivr. Minimac uses slightly different metrics to assess imputation quality (\(R^2\) versus info score) and complete details as to minimac output are available on the Minimac3 Wikipage.

The function, michiganCoxSurv uses a modification of Cox proportional hazard regression from the R library survival:::coxph.fit. Built specifically for genetic data, michiganCoxSurv allows the user to filter on info (\(R^2\)) score (imputation quality metric) and minor allele frequency from the reference panel used for imputation using RefPanelAF as the input arguement for maf.filter. Users are also provided with the sample minor allele frequency (MAF) in the sangerCoxSurv output, which can be used for filtering post analysis.

Samples can be selected for analyses by providing a vector of sample.ids. The output from Sanger imputation server returns the samples as SAMP1, ..., SAMPN, where N is the total number of samples. The sample order corresponds to the sample order in the vcf.file used for imputation. Note, sample order can also be found in the .fam file if genotyping data were initially in .bed, .bim and .fam (PLINK) format prior to conversion to VCF. If no sample list is specified all samples are included in the analyses.

vcf.file <- system.file(package="gwasurvivr",
                        "extdata", 
                        "michigan.chr14.dose.vcf.gz")
pheno.fl <- system.file(package="gwasurvivr",
                        "extdata", 
                        "simulated_pheno.txt")
pheno.file <- read.table(pheno.fl,
                         sep=" ", 
                         header=TRUE,
                         stringsAsFactors = FALSE)
head(pheno.file)
##   ID_1  ID_2 event  time   age DrugTxYes    sex        group
## 1    1 SAMP1     0 12.00 33.93         0   male      control
## 2    2 SAMP2     1  7.61 58.71         1   male experimental
## 3    3 SAMP3     0 12.00 39.38         0 female      control
## 4    4 SAMP4     0  4.30 38.85         0   male      control
## 5    5 SAMP5     0 12.00 43.58         0   male experimental
## 6    6 SAMP6     1  2.60 57.74         0   male      control
# recode sex column and remove first column 
pheno.file$SexFemale <- ifelse(pheno.file$sex=="female", 1L, 0L)
# select only experimental group sample.ids
sample.ids <- pheno.file[pheno.file$group=="experimental",]$ID_2
head(sample.ids)
## [1] "SAMP2"  "SAMP5"  "SAMP7"  "SAMP9"  "SAMP11" "SAMP12"

In this example, we will select samples from the experimental group and will test survival only on these patients. The first column in the pheno.file are sample IDs, which link the phenotype file to the imputation file. We include age, DrugTxYes, and sex in the survival model as covariates.

We perform the analysis using the experimental group to demonstrate how one may want to prepare their data if interested in testing only on a subset of samples (i.e. a case-control study and survival of cases is of interest). Note that how the IDs (sample.ids) need to be a vector of class character. The chunk.size refers to size of each data chunk read in and is defaulted to 10,000 rows. Users can customize that to their needs. The larger the chunk.size the more memory (RAM) required to run the analysis. The recommended chunk.size=10000 and probably should not exceed chunk.size=100000. This does not mean that gwasurvivr is limited to only 100,000 SNPs, it just is how many SNPs are analyzed for each iteration.

By default survival estimates and pvalues for the SNP adjusted for other covariates are outputted (print.covs='only'), however users can select print.covs=all to get the coefficient estimates for covariates included in the model. Depending on the number of covariates included this can add substantially to output file size.

3.1.1 Single SNP analysis

Next we run michiganCoxSurv with the default, print.covs="only", load the results into R and provide descriptions of output by column. We will then run the analysis again using print.covs="all". verbose=TRUE is used for these examples so the function display messages while running.

Use ?michiganCoxSurv for argument specific documentation.

print.covs="only"

michiganCoxSurv(vcf.file=vcf.file,
                covariate.file=pheno.file,
                id.column="ID_2",
                sample.ids=sample.ids,
                time.to.event="time",
                event="event",
                covariates=c("age", "SexFemale", "DrugTxYes"),
                inter.term=NULL,
                print.covs="only",
                out.file="michigan_only",
                r2.filter=0.3,
                maf.filter=0.005,
                chunk.size=100,
                verbose=TRUE,
                clusterObj=NULL)
## Analysis started on 2018-10-30 at 22:30:37
## Covariates included in the models are: age, DrugTxYes, SexFemale
## 52 samples are included in the analysis
## Analyzing chunk 0-100
## Analyzing chunk 100-200
## Analysis completed on 2018-10-30 at 22:30:50
## 0 SNPs were removed from the analysis for not meeting the threshold criteria.
## List of removed SNPs can be found in /tmp/RtmpfCWniv/michigan_only11cb77aeb4d4.snps_removed
## 3 SNPs were analyzed in total
## The survival output can be found at /tmp/RtmpfCWniv/michigan_only11cb77aeb4d4.coxph

Here we load the data and glimpse the first few values in each column outputted from the SNP*interaction survival analyses using print.covs="only".

michigan_only <- read.table("michigan_only.coxph", sep="\t", header=TRUE, stringsAsFactors = FALSE)
str(head(michigan_only))
## 'data.frame':    3 obs. of  21 variables:
##  $ RSID         : chr  "rs34919020" "rs8005305" "rs757545375"
##  $ TYPED        : logi  FALSE FALSE FALSE
##  $ CHR          : int  14 14 14
##  $ POS          : int  19459185 20095842 20097287
##  $ REF          : chr  "C" "G" "A"
##  $ ALT          : chr  "T" "T" "G"
##  $ AF           : num  0.301 0.515 0.52
##  $ MAF          : num  0.301 0.485 0.48
##  $ SAMP_FREQ_ALT: num  0.355 0.517 0.52
##  $ SAMP_MAF     : num  0.355 0.483 0.48
##  $ R2           : num  0.552 0.479 0.481
##  $ ER2          : logi  NA NA NA
##  $ PVALUE       : num  0.713 0.805 0.798
##  $ HR           : num  1.224 0.885 0.881
##  $ HR_lowerCI   : num  0.417 0.333 0.332
##  $ HR_upperCI   : num  3.6 2.35 2.34
##  $ Z            : num  0.368 -0.246 -0.255
##  $ COEF         : num  0.203 -0.123 -0.127
##  $ SE.COEF      : num  0.55 0.498 0.498
##  $ N            : int  52 52 52
##  $ N.EVENT      : int  21 21 21

3.1.2 SNP with covariate interaction

A SNP*covariate interaction can be implemented using the inter.term argument. In this example, we will use DrugTxYes from the covariate file as the covariate we want to test for interaction with the SNP.

print.covs="only"

michiganCoxSurv(vcf.file=vcf.file,
                covariate.file=pheno.file,
                id.column="ID_2",
                sample.ids=sample.ids,
                time.to.event="time",
                event="event",
                covariates=c("age", "SexFemale", "DrugTxYes"),
                inter.term="DrugTxYes",
                print.covs="only",
                out.file="michigan_intx_only",
                r2.filter=0.3,
                maf.filter=0.005,
                chunk.size=100,
                verbose=FALSE,
                clusterObj=NULL)

Here we load the data and glimpse the first few values in each column outputted from the SNP*interaction survival analyses using print.covs="only".

michigan_intx_only <- read.table("michigan_intx_only.coxph", sep="\t", header=TRUE, stringsAsFactors = FALSE)
str(head(michigan_intx_only))
## 'data.frame':    3 obs. of  21 variables:
##  $ RSID         : chr  "rs34919020" "rs8005305" "rs757545375"
##  $ TYPED        : logi  FALSE FALSE FALSE
##  $ CHR          : int  14 14 14
##  $ POS          : int  19459185 20095842 20097287
##  $ REF          : chr  "C" "G" "A"
##  $ ALT          : chr  "T" "T" "G"
##  $ AF           : num  0.301 0.515 0.52
##  $ MAF          : num  0.301 0.485 0.48
##  $ SAMP_FREQ_ALT: num  0.355 0.517 0.52
##  $ SAMP_MAF     : num  0.355 0.483 0.48
##  $ R2           : num  0.552 0.479 0.481
##  $ ER2          : logi  NA NA NA
##  $ PVALUE       : num  0.9017 0.0806 0.0786
##  $ HR           : num  1.15 7.03 7.11
##  $ HR_lowerCI   : num  0.127 0.789 0.799
##  $ HR_upperCI   : num  10.4 62.7 63.2
##  $ Z            : num  0.124 1.747 1.759
##  $ COEF         : num  0.139 1.951 1.961
##  $ SE.COEF      : num  1.12 1.12 1.11
##  $ N            : int  52 52 52
##  $ N.EVENT      : int  21 21 21

3.2 Sanger Imputation Server

Sanger Imputation Server pre-phases typed genotypes using either SHAPEIT or EAGLE, imputes genotypes using PBWT algorithm and outputs a .vcf.gz file for each chromosome. These .vcf.gz files are used as input for gwasurvivr. The function, sangerCoxSurv uses a modification of Cox proportional hazard regression from survival::coxph.fit. Built specifically for genetic data, sangerCoxSurv allows the user to filter on info score (imputation quality metric) and minor allele frequency from the reference panel used for imputation using RefPanelAF as the input arguement for maf.filter. Users are also provided with the sample minor allele frequency in the sangerCoxSurv output.

Samples can be selected for analyses by providing a vector of sample.ids. The output from Sanger imputation server returns the samples as SAMP1, ..., SAMPN, where N is the total number of samples. The sample order corresponds to the sample order in the VCF file used for imputation. Note, sample order can also be found in the .fam file if genotyping data were initially in .bed, .bim and .fam (PLINK) format prior to conversion to VCF. If no sample list is specified, all samples are included in the analyses.

In this example, we will select samples from the experimental group and will test survival only on these patients. The first column in the pheno.file are sample IDs (we will match on these). We include age, DrugTxYes, and sex in the survival model as covariates.

vcf.file <- system.file(package="gwasurvivr",
                        "extdata", 
                        "sanger.pbwt_reference_impute.vcf.gz")
pheno.fl <- system.file(package="gwasurvivr",
                        "extdata", 
                        "simulated_pheno.txt")
pheno.file <- read.table(pheno.fl,
                         sep=" ",
                         header=TRUE,
                         stringsAsFactors = FALSE)
head(pheno.file)
##   ID_1  ID_2 event  time   age DrugTxYes    sex        group
## 1    1 SAMP1     0 12.00 33.93         0   male      control
## 2    2 SAMP2     1  7.61 58.71         1   male experimental
## 3    3 SAMP3     0 12.00 39.38         0 female      control
## 4    4 SAMP4     0  4.30 38.85         0   male      control
## 5    5 SAMP5     0 12.00 43.58         0   male experimental
## 6    6 SAMP6     1  2.60 57.74         0   male      control
# recode sex column and remove first column 
pheno.file$SexFemale <- ifelse(pheno.file$sex=="female", 1L, 0L)
# select only experimental group sample.ids
sample.ids <- pheno.file[pheno.file$group=="experimental",]$ID_2
head(sample.ids)
## [1] "SAMP2"  "SAMP5"  "SAMP7"  "SAMP9"  "SAMP11" "SAMP12"

We perform the analysis using the experimental group to demonstrate how one may want to prepare their data if not initially all samples are patients or cases (i.e. a case-control study and survival of cases is of interest). We also are showing how the IDs (sample.ids) need to be a vector of class character. The chunk.size refers to size of each data chunk read in and is defaulted to 10,000 rows, users can customize that to their needs. The larger the chunk.size the more memory (RAM) required to run the analysis. The recommended chunk.size=10000 and probably should not exceed chunk.size=100000. This does not mean that gwasurvivr is limited to only 100,000 SNPs, it just is how many SNPs are analyzed for each iteration.

By default survival estimates and pvalues for the SNP adjusted for other covariates are outputted (print.covs='only'), however users can select print.covs=all to get the coefficient estimates for covariates included in the model. Depending on the number of covariates included this can add substantially to output file size.

Use ?sangerCoxSurv for argument specific documentation.

3.2.1 Single SNP analysis

Next we run sangerCoxSurv with the default, print.covs="only", load the results into R and provide descriptions of output by column. verbose=TRUE is used for these examples so the function display messages while running.

print.covs="only"

sangerCoxSurv(vcf.file=vcf.file,
              covariate.file=pheno.file,
              id.column="ID_2",
              sample.ids=sample.ids,
              time.to.event="time",
              event="event",
              covariates=c("age", "SexFemale", "DrugTxYes"),
              inter.term=NULL,
              print.covs="only",
              out.file="sanger_only",
              info.filter=0.3,
              maf.filter=0.005,
              chunk.size=100,
              verbose=TRUE,
              clusterObj=NULL)
## Analysis started on 2018-10-30 at 22:31:04
## Covariates included in the models are: age, DrugTxYes, SexFemale
## 52 samples are included in the analysis
## Analyzing chunk 0-100
## Analyzing chunk 100-200
## Analysis completed on 2018-10-30 at 22:31:16
## 0 SNPs were removed from the analysis for not meeting the threshold criteria.
## List of removed SNPs can be found in /tmp/RtmpfCWniv/sanger_only11cbb9a0426.snps_removed
## 3 SNPs were analyzed in total
## The survival output can be found at /tmp/RtmpfCWniv/sanger_only11cbb9a0426.coxph

Here we load the data and glimpse the first few values in each column from the survival analyses.

str(head(sanger_only))
## 'data.frame':    3 obs. of  19 variables:
##  $ RSID         : chr  "rs34919020" "rs8005305" "rs757545375"
##  $ TYPED        : logi  FALSE FALSE FALSE
##  $ CHR          : int  14 14 14
##  $ POS          : int  19459185 20095842 20097287
##  $ REF          : chr  "C" "G" "A"
##  $ ALT          : chr  "T" "T" "G"
##  $ RefPanelAF   : num  0.301 0.515 0.52
##  $ SAMP_FREQ_ALT: num  0.355 0.517 0.52
##  $ SAMP_MAF     : num  0.355 0.483 0.48
##  $ INFO         : num  0.552 0.479 0.481
##  $ PVALUE       : num  0.713 0.805 0.798
##  $ HR           : num  1.224 0.885 0.881
##  $ HR_lowerCI   : num  0.417 0.333 0.332
##  $ HR_upperCI   : num  3.6 2.35 2.34
##  $ Z            : num  0.368 -0.246 -0.255
##  $ COEF         : num  0.203 -0.123 -0.127
##  $ SE.COEF      : num  0.55 0.498 0.498
##  $ N            : int  52 52 52
##  $ N.EVENT      : int  21 21 21

Column names with descriptions from the survival analyses with covariates, specifying the default print.covs="only"

3.2.2 SNP with covariate interaction

A SNP*covariate interaction can be implemented using the inter.term argument. In this example, we will use DrugTxYes from the covariate file as the covariate we want to test for interaction with the SNP.

print.covs="only"

sangerCoxSurv(vcf.file=vcf.file,
              covariate.file=pheno.file,
              id.column="ID_2",
              sample.ids=sample.ids,
              time.to.event="time",
              event="event",
              covariates=c("age", "SexFemale", "DrugTxYes"),
              inter.term="DrugTxYes",
              print.covs="only",
              out.file="sanger_intx_only",
              info.filter=0.3,
              maf.filter=0.005,
              chunk.size=100,
              verbose=TRUE,
              clusterObj=NULL)

3.3 IMPUTE2 Imputation

IMPUTE2 is a genotype imputation and haplotype phasing program (Howie et al 2009). IMPUTE2 outputs 6 files for each chromosome chunk imputed (usually 5 MB in size). Only 2 of these files are required for analyses using gwasurvivr:

  • Genotype file (.impute)
  • Sample file (.sample)

More information can be read about these formats

We are going to load in and pre-process the genetic data and the phenotypic data (covariate.file).

impute.file <- system.file(package="gwasurvivr",
                            "extdata",
                            "impute_example.impute2.gz")
sample.file <- system.file(package="gwasurvivr",
                           "extdata", 
                           "impute_example.impute2_sample")
covariate.file <- system.file(package="gwasurvivr", 
                              "extdata",
                              "simulated_pheno.txt")
covariate.file <- read.table(covariate.file, 
                             sep=" ",
                             header=TRUE,
                             stringsAsFactors = FALSE)
covariate.file$SexFemale <- ifelse(covariate.file$sex=="female", 1L, 0L)
sample.ids <- covariate.file[covariate.file$group=="experimental",]$ID_2

To perform survival analysis using IMPUTE2 the function arguments are very similarto michiganCoxSurv and sangerCoxSurv, however the function now takes a chromosome arguement. This is needed to properly annotate the file output with the chromosome that these SNPs are in. This is purely an artifact of IMPUTE2 and how we leverage GWASTools in this function.

3.3.1 Single SNP analysis

First we will do the analysis with no interaction term, followed by doing the analysis with the interaction term. The recommended output setting for single SNP analysis is print.cov="only".

impute2CoxSurv(impute.file=impute.file,
               sample.file=sample.file,
               chr=14,
               covariate.file=covariate.file,
               id.column="ID_2",
               sample.ids=sample.ids,
               time.to.event="time",
               event="event",
               covariates=c("age", "SexFemale", "DrugTxYes"),
               inter.term=NULL,
               print.covs="only",
               out.file="impute_example_only",
               chunk.size=100,
               maf.filter=0.005,
               exclude.snps=NULL,
               flip.dosage=TRUE,
               verbose=TRUE,
               clusterObj=NULL)
## Covariates included in the models are: age, DrugTxYes, SexFemale
## 52 samples are included in the analysis
## Analysis started on 2018-10-30 at 22:31:28
## Determining number of SNPs and samples...
## Including all SNPs.
## scan.df not given. Assigning scanIDs automatically.
## Reading sample file...
## Reading genotype file...
## Block 1 of 1
## Writing annotation...
## Compressing...
## Clean up the fragments of GDS file:
##     open the file '/tmp/RtmpfCWniv/11cb615cc091.gds' (3.3K)
##     # of fragments: 30
##     save to '/tmp/RtmpfCWniv/11cb615cc091.gds.tmp'
##     rename '/tmp/RtmpfCWniv/11cb615cc091.gds.tmp' (2.4K, reduced: 987B)
##     # of fragments: 14
## ***** Compression time ******
## User:0.056
## System: 0.092
## Elapsed: 0.149
## *****************************
## Analyzing part 1/1...
## 0 SNPs were removed from the analysis for not meeting
## the given threshold criteria or for having MAF = 0
## List of removed SNPs are saved to /tmp/RtmpfCWniv/impute_example_only11cb254ea35e.snps_removed
## In total 3 SNPs were included in the analysis
## The Cox model results output was saved to /tmp/RtmpfCWniv/impute_example_only11cb254ea35e.coxph
## Analysis completed on 2018-10-30 at 22:31:29

Here we load the data and glimpse the first few values in each column output.

impute2_res_only <- read.table("impute_example_only.coxph", sep="\t", header=TRUE, stringsAsFactors = FALSE)
str(head(impute2_res_only))
## 'data.frame':    3 obs. of  17 variables:
##  $ RSID       : chr  "rs34919020" "rs8005305" "rs757545375"
##  $ TYPED      : chr  "---" "---" "---"
##  $ CHR        : int  14 14 14
##  $ POS        : int  19459185 20095842 20097287
##  $ A0         : chr  "C" "G" "A"
##  $ A1         : chr  "T" "T" "G"
##  $ exp_freq_A1: num  0.355 0.517 0.52
##  $ SAMP_MAF   : num  0.355 0.483 0.48
##  $ PVALUE     : num  0.713 0.805 0.798
##  $ HR         : num  1.224 0.885 0.881
##  $ HR_lowerCI : num  0.417 0.333 0.332
##  $ HR_upperCI : num  3.6 2.35 2.34
##  $ Z          : num  0.368 -0.246 -0.255
##  $ COEF       : num  0.203 -0.123 -0.127
##  $ SE.COEF    : num  0.55 0.498 0.498
##  $ N          : int  52 52 52
##  $ N.EVENT    : int  21 21 21

3.3.2 SNP covariate interaction

Now we will perform a SNP*covariate interaction survival analysis using impute2CoxSurv.

impute2CoxSurv(impute.file=impute.file,
               sample.file=sample.file,
               chr=14,
               covariate.file=covariate.file,
               id.column="ID_2",
               sample.ids=sample.ids,
               time.to.event="time",
               event="event",
               covariates=c("age", "SexFemale", "DrugTxYes"),
               inter.term="DrugTxYes",
               print.covs="only",
               out.file="impute_example_intx",
               chunk.size=100,
               maf.filter=0.005,
               flip.dosage=TRUE,
               verbose=FALSE,
               clusterObj=NULL,
               keepGDS=FALSE)
## Determining number of SNPs and samples...
## Including all SNPs.
## scan.df not given. Assigning scanIDs automatically.
## Reading sample file...
## Reading genotype file...
## Block 1 of 1
## Writing annotation...
## Compressing...
## Clean up the fragments of GDS file:
##     open the file '/tmp/RtmpfCWniv/11cb1f96046d.gds' (3.3K)
##     # of fragments: 30
##     save to '/tmp/RtmpfCWniv/11cb1f96046d.gds.tmp'
##     rename '/tmp/RtmpfCWniv/11cb1f96046d.gds.tmp' (2.4K, reduced: 987B)
##     # of fragments: 14
## ***** Compression time ******
## User:0.096
## System: 0.104
## Elapsed: 0.203
## *****************************

Here we load the data and glimpse the first few values in each column outputted from the SNP*interaction survival analyses using print.covs="only".

impute2_res_intx <- read.table("impute_example_intx.coxph", sep="\t", header=TRUE, stringsAsFactors = FALSE)
str(head(impute2_res_intx))
## 'data.frame':    3 obs. of  17 variables:
##  $ RSID       : chr  "rs34919020" "rs8005305" "rs757545375"
##  $ TYPED      : chr  "---" "---" "---"
##  $ CHR        : int  14 14 14
##  $ POS        : int  19459185 20095842 20097287
##  $ A0         : chr  "C" "G" "A"
##  $ A1         : chr  "T" "T" "G"
##  $ exp_freq_A1: num  0.355 0.517 0.52
##  $ SAMP_MAF   : num  0.355 0.483 0.48
##  $ PVALUE     : num  0.9017 0.0806 0.0786
##  $ HR         : num  1.15 7.03 7.11
##  $ HR_lowerCI : num  0.127 0.789 0.799
##  $ HR_upperCI : num  10.4 62.7 63.2
##  $ Z          : num  0.124 1.747 1.759
##  $ COEF       : num  0.139 1.951 1.961
##  $ SE.COEF    : num  1.12 1.12 1.11
##  $ N          : int  52 52 52
##  $ N.EVENT    : int  21 21 21

4 plinkCoxSurv

5 Batch Examples

5.1 Batch Example sangerCoxSurv

Batch jobs for multiple analyses and different subsets are easy to implement using gwasurvivr. These types of analyses should be reserved for usage on a UNIX-based high performance computing cluster. This is facilitated by the package batch, which can internalize R variables from bash. First write an R script (e.g. mysurvivalscript.R) to pass in bash.

Note: it is important to refer to the Getting Started part of the manual for packages that need to be installed for this

## mysurvivalscript.R
library(gwasurvivr)
library(batch)

parseCommandArgs(evaluate=TRUE) # this is loaded in the batch package

options("gwasurvivr.cores"=4)

vcf.file <- system.file(package="gwasurvivr",
                        "extdata", 
                        vcf.file)
pheno.fl <- system.file(package="gwasurvivr",
                        "extdata", 
                        pheno.file)
pheno.file <- read.table(pheno.fl, 
                         sep=" ",
                         header=TRUE, 
                         stringsAsFactors = FALSE)
# recode sex column and remove first column 
pheno.file$SexFemale <- ifelse(pheno.file$sex=="female", 1L, 0L)
# select only experimental group sample.ids
sample.ids <- pheno.file[pheno.file$group=="experimental",]$ID_2
## -- unlist the covariates 
## (refer below to the shell script as to why we are doing this)
covariates <- unlist(sapply(covariates, strsplit, "_", 1, "[["))

sangerCoxSurv(vcf.file=vcf.file,
              covariate.file=pheno.file,
              id.column="ID_2",
              sample.ids=sample.ids,
              time.to.event=time,
              event=event,
              covariates=covariates,
              inter.term=NULL,
              print.covs="only",
              out.file=out.file,
              info.filter=0.3,
              maf.filter=0.005,
              chunk.size=100,
              verbose=TRUE,
              clusterObj=NULL)

Now we can run a shell script. This can be used well with manifest files to set up multiple runs with different outcomes and different subsets of samples. We define a manifest file has columns that corresond to the functions that the user wants to pass and each row is a separate analysis that a user may want to run. The covariates are separated by an underscore ("_"). This is so it can be passed properly, and also why we used str_split to split the covariates.

#!/bin/bash
DIRECTORY=/path/to/dir/impute_chr

module load R 

R --script ${DIRECTORY}/survival/code/mysurvivalscript.R -q --args \
        vcf.file ${DIRECTORY}/sanger.pbwt_reference_impute.vcf.gz \
        pheno.file ${DIRECTORY}/phenotype_data/simulated_pheno.txt \
        covariates DrugTxYes_age_SexFemale\
        time.to.event time \
        event event \
        out.file ${DIRECTORY}/survival/results/sanger_example_output

The file paths above are completely arbitrary and were just used as an example of how file structure may be and where desirable output would be stored.

5.2 Batch Example impute2CoxSurv

Exactly the same as for sangerCoxSurv but this time with the input arguments for impute2CoxSurv. See ?impute2CoxSurv for help

5.3 Batch Example michiganCoxSurv

Exactly the same as for sangerCoxSurv but this time with the input arguments for michiganCoxSurv. See ?michiganCoxSurv for help

Session info

Here is the output of sessionInfo() on the system that this document was compiled:

## R version 3.5.1 Patched (2018-07-12 r74967)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 16.04.5 LTS
## 
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.8-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.8-bioc/R/lib/libRlapack.so
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=C              
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] gwasurvivr_1.0.0 knitr_1.20       BiocStyle_2.10.0
## 
## loaded via a namespace (and not attached):
##  [1] nlme_3.1-137                bitops_1.0-6               
##  [3] matrixStats_0.54.0          bit64_0.9-7                
##  [5] httr_1.3.1                  progress_1.2.0             
##  [7] rprojroot_1.3-2             GenomeInfoDb_1.18.0        
##  [9] tools_3.5.1                 backports_1.1.2            
## [11] R6_2.3.0                    rpart_4.1-13               
## [13] DBI_1.0.0                   BiocGenerics_0.28.0        
## [15] mgcv_1.8-25                 SNPRelate_1.16.0           
## [17] jomo_2.6-4                  nnet_7.3-12                
## [19] DNAcopy_1.56.0              prettyunits_1.0.2          
## [21] tidyselect_0.2.5            bit_1.1-14                 
## [23] compiler_3.5.1              quantreg_5.36              
## [25] Biobase_2.42.0              mice_3.3.0                 
## [27] SparseM_1.77                DelayedArray_0.8.0         
## [29] sandwich_2.5-0              rtracklayer_1.42.0         
## [31] bookdown_0.7                lmtest_0.9-36              
## [33] quantsmooth_1.48.0          stringr_1.3.1              
## [35] digest_0.6.18               Rsamtools_1.34.0           
## [37] minqa_1.2.4                 GWASExactHW_1.01           
## [39] rmarkdown_1.10              XVector_0.22.0             
## [41] pkgconfig_2.0.2             htmltools_0.3.6            
## [43] lme4_1.1-18-1               BSgenome_1.50.0            
## [45] highr_0.7                   rlang_0.3.0.1              
## [47] RSQLite_2.1.1               bindr_0.1.1                
## [49] zoo_1.8-4                   BiocParallel_1.16.0        
## [51] dplyr_0.7.7                 VariantAnnotation_1.28.0   
## [53] RCurl_1.95-4.11             magrittr_1.5               
## [55] GenomeInfoDbData_1.2.0      Matrix_1.2-14              
## [57] Rcpp_0.12.19                S4Vectors_0.20.0           
## [59] stringi_1.2.4               yaml_2.2.0                 
## [61] GWASTools_1.28.0            MASS_7.3-51                
## [63] SummarizedExperiment_1.12.0 zlibbioc_1.28.0            
## [65] grid_3.5.1                  blob_1.1.1                 
## [67] parallel_3.5.1              mitml_0.3-6                
## [69] crayon_1.3.4                lattice_0.20-35            
## [71] Biostrings_2.50.0           splines_3.5.1              
## [73] GenomicFeatures_1.34.0      hms_0.4.2                  
## [75] pillar_1.3.0                GenomicRanges_1.34.0       
## [77] logistf_1.23                biomaRt_2.38.0             
## [79] gdsfmt_1.18.0               stats4_3.5.1               
## [81] pan_1.6                     XML_3.98-1.16              
## [83] glue_1.3.0                  evaluate_0.12              
## [85] BiocManager_1.30.3          nloptr_1.2.1               
## [87] MatrixModels_0.4-1          purrr_0.2.5                
## [89] tidyr_0.8.2                 assertthat_0.2.0           
## [91] xfun_0.4                    broom_0.5.0                
## [93] survival_2.43-1             tibble_1.4.2               
## [95] GenomicAlignments_1.18.0    AnnotationDbi_1.44.0       
## [97] memoise_1.1.0               IRanges_2.16.0             
## [99] bindrcpp_0.2.2

References

  1. Terry M. Therneau and Patricia M. Grambsch (2000). Modeling Survival Data: Extending the Cox Model. Springer, New York. ISBN 0-387-98784-3.

  2. Martin Morgan, Valerie Obenchain, Jim Hester and Hervé Pagès (2017). SummarizedExperiment: SummarizedExperiment container. R package version 1.6.3.

  3. Gogarten SM, Bhangale T, Conomos MP, Laurie CA, McHugh CP, Painter I, Zheng X, Crosslin DR, Levine D, Lumley T, Nelson SC, Rice K, Shen J, Swarnkar R, Weir BS and Laurie CC (2012). “GWASTools: an R/Bioconductor package for quality control and analysis of genome-wide association studies.” Bioinformatics, 28(24), pp. 3329-3331. doi: 10.1093/bioinformatics/bts610.

  4. B. N. Howie, P. Donnelly and J. Marchini (2009) A flexible and accurate genotype imputation method for the next generation of genome-wide association studies. PLoS Genetics 5(6): e1000529

  5. Das S, Forer L, Schönherr S, Sidore C, Locke AE, Kwong A, Vrieze S, Chew EY, Levy S, McGue M, Schlessinger D, Stambolian D, Loh PR, Iacono WG, Swaroop A, Scott LJ, Cucca F, Kronenberg F, Boehnke M, Abecasis GR, Fuchsberger C. Next-generation genotype imputation service and methods. Nature Genetics 48, 1284–1287 (2016). 27571263

  6. Efficient haplotype matching and storage using the Positional Burrows-Wheeler Transform (PBWT)“, Richard Durbin Bioinformatics 30:1266-72 (2014).