R, ggplot, separate mean by range of x value

Question

I have a set of data looks like this

  CHROM   POS GT DIFF
1 chr01 14653 CT 254
2 chr01 14907 AG 254
3 chr01 14930 AG 23
4 chr01 15190 GA 260
5 chr01 15211 TG 21
6 chr01 16378 TC 1167

Where POS range from 1xxxx to 1xxxxxxx. And CHROM is a categorical variable that contains values of "chr01" to "chr22" and "chrX".

I want to plot a scatterplot:

• y(DIFF) vs. X(POS)
• having panels separated by CHROM
• grouped by GT (different colors by GT)

I'm creating a ggplot with running average (though not time series data).

What I want is to get average for every 1,000,000 range of POS by GT.

For example,

for x in range(1 ~ 1,000,000) , DIFF average = _____

for x in range(1,000,001 ~ 2,000,000), DIFF average = _____

and I want to plot horizontal lines on the ggplot (coloured by GT).

#

What I have so far before apply your function:

After apply your function:

I tried to apply your solution to what I already have, here are some problems:

• There are different panels, so the mean values are different for different panel, but when I apply your code, the horizontal mean lines are all identical to the first panel.
• I'm having different ranges for x-axis, so when apply your function, it automatically fills out the extra range with the previous horizontal mean line

Here is my code before:

ggplot(data1, aes(x=POS,y=DIFF,colour=GT)) +
  geom_point() +
  facet_grid(~ CHROM,scales="free_x",space="free_x") + 
  theme(strip.text.x = element_text(size=40),
        strip.background = element_rect(color='lightblue',fill='lightblue'),
        legend.position="top",
        legend.title = element_text(size=40,colour="darkblue"),
        legend.text = element_text(size=40),
        legend.key.size = unit(2.5, "cm")) +
  guides(fill = guide_legend(title.position="top",
                             title = "Legend:GT='REF'+'ALT'"),
         shape = guide_legend(override.aes=list(size=10))) +
  scale_y_log10(breaks=trans_breaks("log10", function(x) 10^x, n=10)) + 
  scale_x_continuous(breaks = pretty_breaks(n=3))

Answer 1

This should get you started:

# It saves a lot of headaches to just make factors as you need them
options(stringsAsFactors = FALSE)



library(ggplot2)
library(plyr)

# Here's some made-up data - it always helps if you can post a subset of
# your real data, though. The dput() function is really useful for that.
dat <- data.frame(POS = seq(1, 1e7, by = 1e4))


# Add random GT value
dat$GT <- sample(x = c("CT", "AG", "GA", "TG", "TC"),
                 size = nrow(dat),
                 replace = TRUE)

# Group by millions - there are several ways to do this that I can 
# never remember, but here's a simple way to split by millions
dat$POSgroup <- floor(dat$POS / 1e6)


# Add an arbitrary DIFF value
dat$DIFF <- rnorm(n = nrow(dat),
                  mean = 200 * dat$POSgroup,
                  sd = 300)



# Aggregate the data by GT and POS-group
# Ideally, you'd do this inside of the plot using stat_summary,
# but I couldn't get that to work. Using two datasets in a plot 
# is okay, though.
datsum <- ddply(dat, .var = "POSgroup", .fun = function(x) {

    # Calculate the mean DIFF value for each GT group in this POSgroup
    meandiff <- ddply(x, .var = "GT", .fun = summarise, ymean = mean(DIFF))
                
    # Add the center of the POSgroup range as the x position
    meandiff$center <- (x$POSgroup[1] * 1e6) + 0.5e6

    # Return the results
    meandiff

})


# On the plot, these results will be grouped by both POS and GT - but
# ggplot will only accept one vector for grouping. So make a combination.
datsum$combogroup <- paste(datsum$GT, datsum$POSgroup)


# Plot it
ggplot() +

    # First, a layer for the points themselves
    # Large numbers of points can get pretty slow - you might try getting
    # the plot to work with a subsample (~1000) and then add in the rest of
    # your data
    geom_point(data = dat, 
               aes(x = POS, y = DIFF, color = as.factor(GT))) +

    # Then another layer for the means. There are a variety of geoms you could
    # use here, but crossbar with ymin and ymax set to the group mean
    # is a simple one
    geom_crossbar(data = datsum, aes(x = center, 
                                     y = ymean, 
                                     ymin = ..y.., 
                                     ymax = ..y.., 
                                     color = as.factor(GT),
                                     group = combogroup),
                  size = 1) +


    # Some other niceties
    scale_x_continuous(breaks = seq(0, 1e7, by = 1e6)) +
    labs(x = "POS", y = "DIFF", color = "GT") +
    theme_bw()

Which results in this: