view scater-pca-filter.R @ 1:946179ef029c draft default tip

"planemo upload for repository https://github.com/galaxyproject/tools-iuc/tree/master/tools/scater commit 61f3899168453092fd25691cf31871a3a350fd3b"
author iuc
date Tue, 03 Sep 2019 14:28:53 -0400
parents 87757f7b9974
children
line wrap: on
line source

#!/usr/bin/env Rscript
#Filters a SingleCellExperiment object, using PCA on the following metrics:
# "pct_counts_top_100_features"
# "total_features"
# "pct_counts_feature_control"
# "total_features_feature_control"
# "log10_total_counts_endogenous"
# "log10_total_counts_feature_control"

# Load optparse we need to check inputs
library(optparse)
library(workflowscriptscommon)
library(LoomExperiment)
library(scater)
library(mvoutlier)

# parse options
option_list = list(
  make_option(
    c("-i", "--input-loom"),
    action = "store",
    default = NA,
    type = 'character',
    help = "A SingleCellExperiment object file in Loom format."
  ),
  make_option(
    c("-o", "--output-loom"),
    action = "store",
    default = NA,
    type = 'character',
    help = "File name in which to store the SingleCellExperiment object in Loom format."
  )
)

opt <- wsc_parse_args(option_list, mandatory = c('input_loom', 'output_loom'))

# Check parameter values

if ( ! file.exists(opt$input_loom)){
  stop((paste('File', opt$input_loom, 'does not exist')))
}

# Input from Loom format

scle <- import(opt$input_loom, format='loom', type='SingleCellLoomExperiment')
print(paste("Starting with", ncol(scle), "cells and", nrow(scle), "features."))

# Run PCA on data and detect outliers
scle <- runPCA(scle, use_coldata = TRUE, detect_outliers = TRUE)

# Filter out outliers
scle <- scle[, !scle$outlier]

print(paste("Ending with", ncol(scle), "cells and", nrow(scle), "features."))

# Output to a Loom file
if (file.exists(opt$output_loom)) {
  file.remove(opt$output_loom)
}
export(scle, opt$output_loom, format='loom')