filter_zero_var_genes.R

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

# read data from causton heat expression data set
# see groups 4 and 9 data/focal papers
causton <- read.csv("causton-2001-heat-expression.csv")

# Take a look at data. You'll see that the genes are in rows, 
# and the time points and gene names are in columns
head(causton)

# Reshape data by
causton.long <- 
  causton %>%
  gather(time, expression, -ORF, -Gene) %>%
  arrange(ORF)  # sort by ORF

# look at the data after reshaping
head(causton.long)

# Calculate the variance by gene
causton.var <-
  causton.long %>%
  group_by(ORF) %>%
  summarize(var = var(expression, na.rm = TRUE))

# Find the ORFs with non-zero variance
non.zero.var.ORFs <-
  causton.var %>%
  filter(var != 0) %$%  # NOTE use of %$% operator from magrittr pkg
  ORF

# only keep genes with non-zero variance
causton.trim <-
  causton.long %>%
  filter(ORF %in% non.zero.var.ORFs) 

# spread to "wide" format appropriate for calculating correlations, with
# genes in columns, time points in row, after dropping Gene name column
causton.wide <-
  causton.trim %>%
  select(-Gene) %>%  # drop the Gene name column
  spread(ORF, expression)

# calculating correlations
causton.cor <- 
  causton.wide %>%
  select(-time) %>%  # drop the time column before calculating correlations
  cor(use = "pairwise.complete.obs")