## ----setup, include=FALSE--------------------------------------------------
knitr::opts_chunk$set(
fig.path='figure/graphics-',
cache.path='cache/graphics-',
fig.align='center',
external=TRUE,
echo=TRUE,
warning=FALSE
# fig.pos="H"
)
## ---- eval=TRUE, echo=FALSE, message=FALSE, warning=FALSE------------------
rTASSEL::startLogger()
## ---- eval=FALSE, echo=TRUE------------------------------------------------
# if (!require("devtools")) install.packages("devtools")
# devtools::install_bitbucket(repo = "bucklerlab/rtassel", ref = "master")
## ---- eval=FALSE, echo=TRUE------------------------------------------------
# library(rTASSEL)
## ---- eval=FALSE, echo=TRUE------------------------------------------------
# rTASSEL::startLogger(fullPath = NULL, fileName = NULL)
## ---- eval=FALSE, echo=TRUE------------------------------------------------
# options(java.parameters = c("-Xmx<memory>", "-Xms<memory>"))
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
# Load hapmap data
genoPathHMP <- system.file(
"extdata",
"mdp_genotype.hmp.txt",
package = "rTASSEL"
)
genoPathHMP
# Load VCF data
genoPathVCF <- system.file(
"extdata",
"maize_chr9_10thin40000.recode.vcf",
package = "rTASSEL"
)
genoPathVCF
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
# Load in hapmap file
tasGenoHMP <- rTASSEL::readGenotypeTableFromPath(
path = genoPathHMP
)
# Load in VCF file
tasGenoVCF <- rTASSEL::readGenotypeTableFromPath(
path = genoPathVCF
)
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasGenoHMP
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
class(tasGenoHMP)
slotNames(tasGenoHMP)
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasGenoHMP@jGenotypeTable
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
# Read from phenotype path
phenoPath <- system.file("extdata", "mdp_traits.txt", package = "rTASSEL")
phenoPath
# Load into rTASSEL `TasselGenotypePhenotype` object
tasPheno <- rTASSEL::readPhenotypeFromPath(
path = phenoPath
)
# Inspect object
tasPheno
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
# Create phenotype data frame
phenoDF <- read.table(phenoPath, header = TRUE)
colnames(phenoDF)[1] <- "Taxon"
# Inspect first few rows
head(phenoDF)
# Load into rTASSEL `TasselGenotypePhenotype` object
tasPhenoDF <- rTASSEL::readPhenotypeFromDataFrame(
phenotypeDF = phenoDF,
taxaID = "Taxon",
attributeTypes = NULL
)
# Inspect new object
tasPhenoDF
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasGenoPheno <- rTASSEL::readGenotypePhenotype(
genoPathOrObj = tasGenoHMP,
phenoPathDFOrObj = tasPheno
)
tasGenoPheno
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasGenoPhenoDF <- rTASSEL::readGenotypePhenotype(
genoPathOrObj = genoPathHMP,
phenoPathDFOrObj = phenoDF,
taxaID = "Taxon",
attributeTypes = NULL
)
tasGenoPhenoDF
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasSumExp <- rTASSEL::getSumExpFromGenotypeTable(
tasObj = tasGenoPheno
)
tasSumExp
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
SummarizedExperiment::colData(tasSumExp)
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
SummarizedExperiment::rowData(tasSumExp)
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
SummarizedExperiment::rowRanges(tasSumExp)
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasExportPhenoDF <- rTASSEL::getPhenotypeDF(
tasObj = tasGenoPheno
)
tasExportPhenoDF
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasGenoPhenoFilt <- rTASSEL::filterGenotypeTableSites(
tasObj = tasGenoPheno,
siteMinCount = 150,
siteMinAlleleFreq = 0.05,
siteMaxAlleleFreq = 1.0
)
tasGenoPhenoFilt
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasGenoPheno
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasKin <- rTASSEL::kinshipMatrix(tasObj = tasGenoPheno)
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
# Get full R matrix
tasKinRMat <- rTASSEL::kinshipToRMatrix(tasKin)
# Inspect the first 5 rows and columns
tasKinRMat[1:5, 1:5]
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
tasDist <- rTASSEL::distanceMatrix(tasObj = tasGenoPheno)
## ---- eval=TRUE, echo=TRUE-------------------------------------------------
# Get full R matrix
tasDistRMat <- rTASSEL::distanceToRMatrix(tasDist)
# Inspect the first 5 rows and columns
tasDistRMat[1:5, 1:5]
## ---- echo=TRUE, eval=TRUE-------------------------------------------------
# Read in phenotype data
phenoPathCov <- system.file("extdata", "mdp_phenotype.txt", package = "rTASSEL")
tasPhenoCov <- rTASSEL::readPhenotypeFromPath(phenoPathCov)
# Calculate BLUEs
tasBLUE <- rTASSEL::assocModelFitter(
tasObj = tasPhenoCov,
formula = . ~ ., # <- All data is used!
fitMarkers = FALSE,
kinship = NULL,
fastAssociation = FALSE
)
# Return BLUE output
tasBLUE
## ---- echo=TRUE, eval=TRUE-------------------------------------------------
# Calculate GLM
tasGLM <- rTASSEL::assocModelFitter(
tasObj = tasGenoPheno, # <- our prior TASSEL object
formula = list(EarHT, dpoll) ~ ., # <- only EarHT and dpoll are ran
fitMarkers = TRUE, # <- set this to TRUE for GLM
kinship = NULL,
fastAssociation = FALSE
)
# Return GLM output
tasGLM
## ---- echo=TRUE, eval=TRUE-------------------------------------------------
# Calculate MLM
tasMLM <- rTASSEL::assocModelFitter(
tasObj = tasGenoPheno, # <- our prior TASSEL object
formula = EarHT ~ ., # <- run only EarHT
fitMarkers = TRUE, # <- set this to TRUE for GLM
kinship = tasKin, # <- our prior kinship object
fastAssociation = FALSE
)
# Return GLM output
tasMLM
## ---- echo=TRUE, eval=TRUE-------------------------------------------------
# Read data - need only non missing data!
phenoPathFast <-system.file(
"extdata",
"mdp_traits_nomissing.txt",
package = "rTASSEL"
)
# Creat rTASSEL object - use prior TASSEL genotype object
tasGenoPhenoFast <- rTASSEL::readGenotypePhenotype(
genoPathOrObj = tasGenoHMP,
phenoPathDFOrObj = phenoPathFast
)
# Calculate MLM
tasFAST <- rTASSEL::assocModelFitter(
tasObj = tasGenoPhenoFast, # <- our prior TASSEL object
formula = . ~ ., # <- run all of the phenotype data
fitMarkers = TRUE, # <- set this to TRUE for GLM
kinship = NULL,
fastAssociation = TRUE # <- set this to TRUE for fast assoc.
)
# Return GLM output
tasFAST
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.