Simple Graphics for Gene Expression

In the 1980s, the National Cancer Institute developed a set of 60 cancer cell lines, called NCI60. The original purpose was for screening potential anti-cancer drugs. A recent overview in Nature describes some of the ongoing work with these cell lines.

Here you will examine gene expression in these cell lines using data described in Staunton et al.. More than 41,000 probes were used for each of the 60 cell lines.

For convenience, the data are provided by the DCI package in two data tables NCI60 and NCI60cells.1

NCI60 is somewhat large — 41,078 probes by 60 cell lines. Each of those 2,454,680 entries is a measure of how much a particular gene was expressed in one cell line.

In this exercise, you’re going to look at just one of the probes. The question you will make a graphic to address is:

Is the probe’s gene expressed in different amounts across different types of cancer?

This activity is intended to be about the construction of informative graphics, not the mechanics of transfiguring data. To save time, the following set of commands should be copied into your report, so that you can execute them to create glyph-ready data.

You should be able to read and understand the commands, but you don’t have to construct them from scratch.2

probeSelected <- "[Your probe name, e.g. DZIP1L, goes here.]"
Narrow <-
  NCI60 %>% 
  tidyr::gather( cellLine, expression, -Probe ) %>%
  mutate( cellLine=as.character(cellLine) )
JustTheTypes <- 
  NCI60cells %>%
  mutate( cellLine=as.character(cellLine) ) %>% 
  mutate( cellLine=gsub("\\:",".", cellLine)) %>%
  select(cellLine, tissue )
Narrow <- inner_join( Narrow, JustTheTypes )
MyProbe <- 
  Narrow %>% 
  filter( Probe==probeSelected )

The Narrow and MyProbe data frames are in glyph-ready form.

We’re ready for a simple graphic: Make a scatter plot of the expression versus tissue type:

ggplot( MyProbe, aes(x=tissue, y=expression)) +
  geom_point() + 
  theme(axis.text.x=element_text(angle=60,hjust=1))

plot of chunk unnamed-chunk-7

It’s hard to see, at a glance, the differences between expressions in the various tissue types. Therefore, this is not a good graphic.

It’s very common in the professional literature to display data like these as a bar chart. It might look like this:

SimpleStats <-
  MyProbe %>%
  group_by( tissue ) %>%
  summarise( mid=mean( expression ))
ggplot( SimpleStats, aes(x=tissue, y=mid )) + 
  geom_bar( stat="identity" ) +
  theme(axis.text.x=element_text(angle=60,hjust=1))

plot of chunk unnamed-chunk-8

Before going on, decide what you like and don’t like about this plot. Write it down

Then continue …

… but not before you’ve written down your opinions …

… Really!

… Are you ready now?

… You’ve really written something?

… Then go on.

There are several bad features of this plot:

Here are several suggestions for improving the graphic:

  1. Lighten up on the color. Perhaps alpha=.2. Or perhaps a dot plot rather than a bar chart.

  2. Reorder the tissue types.

    One way to do this is to pick a quantity that you want to dictate the order of groups. For instance, the mean expression.
    • Create a data table of that quantity for each group (as with SimpleStats).

    • Reorder the categories based on the quantity:

      SimpleStats2 <-
  SimpleStats %>%
  mutate( tissue=reorder( tissue, mid ) )
      Use disc() to order the other way.

  3. Show a statistical measure of the variation.

    Without going into details of statistical method, a very common way to present the imprecision in an estimated quantity is with a “standard error.” For quantities such as the mean, the standard error has a simple form (but not necessarily a form that will be obvious to you). Here’s the calculation:

    SimpleStats3 <-
  MyProbe %>%
  group_by( tissue ) %>%
  summarise( mid=mean( expression ),
         se=sd(expression)/sqrt(n()) )
  4. Show the expression values for individual cases.

  5. Use a different modality, e.g. a dot plot, a box-and-whiskers plot (with notch=TRUE).

The conventional solution is not the best: show the bars along with the measure of precision.

plot of chunk unnamed-chunk-11

Statisticians refer to such things as “dynamite plots,” in a way intended to be pejorative. You can find a better way to present these data.

Explication of the Data Transfiguration

The NCI60 table is in wide format: each of the cell lines is its own variable. You’re going to turn it into a narrow format.

  1. Look at the variable names in NCI60. How many are there?

  2. Run this command:

    Narrow <-
  NCI60 %>% 
  gather( cellLine, expression, -Probe ) %>%
  mutate( cellLine=as.character(cellLine) )

Look at a small random set of the cases in Narrow. How many variables are there now?

Add in the cell types (with some technical corrections to make names match)

JustTheTypes <- 
  NCI60cells %>%
  mutate( cellLine=as.character(cellLine) ) %>% 
  mutate( cellLine=gsub("\\:",".", cellLine)) %>%
  select(cellLine, tissue )

Now, add the kind of cell line the each case belongs to. This information is in the NCI60cells table. The variables already have matching names.

Narrow <- inner_join( Narrow, JustTheTypes )

Pick one of these probes to look at, it doesn’t matter which one.3 A_32_P33263 CP FSD1 A_32_P52227 A_23_P135646 FLJ38359 ZNF682 A_24_P936145 ALS2CR4 DZIP1L LOC492303 ZNF277P CBFA2T3 MUC13 A_32_P168727

Using your probe, filter out only the cases involving that probe, for instance:

The result, MyProbe has 60 to 600 cases. It should look like this:

##         Probe      cellLine expression tissue
## 1 A_32_P33263       BR.MCF7      -1.58 Breast
## 2 A_32_P33263 BR.MDA_MB_231      -5.47 Breast
## 3 A_32_P33263     BR.HS578T      -6.18 Breast
## 4 A_32_P33263     BR.BT_549      -5.91 Breast

Now, MyProbe will look like this:

##         Probe      cellLine expression tissue
## 1 A_32_P33263       BR.MCF7      -1.58 Breast
## 2 A_32_P33263 BR.MDA_MB_231      -5.47 Breast
## 3 A_32_P33263     BR.HS578T      -6.18 Breast

  1. NCI60 and NCI60cells are lightly re-organized from the form provided by Staunton et al..↩

  2. A few lines involve data cleaning, which you have not yet studied. Don’t worry about the lines where as.character() or gsub() appear.↩

  3. In reality, the choice of probe makes a great deal of difference. These particularly probes were chosen based on a screening test. In an upcoming part of the course, you’ll look at how to deal with multiple variables, such as the 41,078 variables here.↩