filter_genes_by_var.R

library(tidyr)
library(dplyr)
library(magrittr)
library(ggplot2)

# load data in "wide" format (genes in columns)
spellman <- read.csv("spellman-reformated.csv")

# restructure in "long" format
spellman.long <- gather(spellman, gene, expression, -expt, -time)

# group by gene, and calculate variance of expression for each gene
# removing any NA values
spellman.var <- 
  spellman.long %>%
  group_by(gene) %>%
  summarize(var = var(expression, na.rm=TRUE))

## make a quick histogram of the variances
ggplot(spellman.var, aes(var)) + geom_histogram()

# Let's find the cutotff to remove the bottom 1/3 least variable genes
# find the cutoff point corresponding to the 30% percentile of
# the variances
cutoff <- quantile(spellman.var$var, 0.33)

# Get genes whose variance is greater than cutoff
genes.of.interest <- spellman.var %>% filter(var >= cutoff) %$% gene
# the %$% operator comes from the magrittr package 
# see https://github.com/tidyverse/magrittr

# create reduced "wide" data frame
spellman.reduced <- 
  spellman %>%
  select(one_of(genes.of.interest))

# equivalent for our "long" data frame
spellman.reduced.long <-
  spellman.long %>%
  filter(gene %in% genes.of.interest)

# original dimension of spellman data
dim(spellman)

# dimension of 
dim(spellman.reduced)