stringr::str_trim(" Removing whitespaces at both ends ")[1] "Removing whitespaces at both ends"Yeast transcriptomics
Objective
Objective: summarise a large transcriptomics study using linear regressions. Experience how wrangling real data can actually take 80% of the analysis job.
Yeast data
In 2008, Brauer et al. used microarrays to test the effect of starvation on the growth rate of yeast. For example, they limit the yeast’s supply of glucose (sugar to metabolize into energy), leucine (an essential amino acid), or of ammonia (a source of nitrogen) and assess how yeast cells reacted to this stress, how they adapt certain genes expression. Brauer et al, tested several growth rates in their chemostat, which means that the lower the growth rate is, the more severe the starvation for a nutrient is.
Retreive the data
Download Brauer2008_DataSet1.tds inside a data sub-folder you should create.
Load the Brauer2008_DataSet1.tds file as a tibble named original_data. This is the exact data that was published with the paper (though for some reason the link on the journal’s page is broken). It thus serves as a good example of tidying a biological dataset “found in the wild”.
Tidying the data
Have a look at the dataset. Is the data “tidy”?
Many variables are stored in one column NAME
- Gene name e.g. SFB2. Note that not all genes have a name.
- Biological process e.g. “proteolysis and peptidolysis”
- Molecular function e.g. “metalloendopeptidase activity”
- Systematic ID e.g. YNL049C. Unlike a gene name, every gene in this dataset has a systematic ID.
- Another ID number e.g.
1082129. We don’t know what this number means, and it’s not annotated in the paper. Oh, well, we will discard it eventually.
Use the appropriate tidyr function to separate these values and generate a column for each variable. Save as cleaned_data
Tip
Preferred names for new columns are;
- “name”
- “BP”
- “MF”
- “systematic_name”
- “number”
Once you separated the variables delimited by two “||”, check closer the new values: You will see that in columns like systematic_name, BP etc values are surrounded by white-spaces which might be inconvenient during the subsequent use.
For example, on the data below
# A tibble: 5,537 x 44
GID YORF name BP MF systematic_name number GWEIGHT G0.05 G0.1 [...]
<chr> <chr> <chr> <chr> <chr> <chr> <chr> <int> <dbl> <dbl> [...]
1 GENE1… A_06… "SFB… " ER… " mo… " YNL049C " " 108… 1 -0.24 -0.13 [...]
2 GENE4… A_06… "" " bi… " mo… " YNL095C " " 108… 1 0.28 0.13 [...]
3 GENE4… A_06… "QRI… " pr… " me… " YDL104C " " 108… 1 -0.02 -0.27 [...]
[...]the test systematic_name == "YNL049C" is FALSE while systematic_name == " YNL049C " is TRUE
Remove the white spaces that start and end strings in the columns name to number. Save as cleaned_ws_data
Tip
dplyr allows us to apply a function to selected columns using across(). To remove these white-spaces, stringr provides a function called str_trim(). Let’s test how the function works:
We are not going to use every column of the dataframe. Remove the unnecessary columns: number, GID, YORF and GWEIGHT. Save as cleaned_ws_data
Do you think that our dataset is now tidy?
Column names must not contain values, pivot the tibble so each column represents a variable. Save as cleaned_data_melt
Tip
At this point we are storing the sample name (will contain G0.05 …) as a new column sample and values in a column named expression.
Observe the sample column, print the unique values of this column.
We are again facing the problem that two variables are stored in a single column. The nutrient (G, N etc.) is the first character, then the growth rate.
Use the same function as before to split the sample column into two variables nutrient and rate. Assign the name cleaned_data_melt_sp
Tip
Use separate() and the appropriate delimitation in sep. Consider using the convert argument. It allows to convert strings to number when relevant like here.
Turn nutrient letters into more comprehensive words
Right now, the nutrients are designed by a single letter. It would be nice to have the full word instead. One could use a full mixture of if and else such as if_else(nutrient == "G", "Glucose", if_else(nutrient == "L", "Leucine", etc ...)) But, that would be cumbersome.
Using the following correspondences and dplyr::recode(), recode all nutrient names with their full explicit names. Save as cleaned_data_melt_nut
Here is the list of the correspondences:
G = "Glucose", L = "Leucine", P = "Phosphate",
S = "Sulfate", N = "Ammonia", U = "Uracil"Cleaning up missing data
Two variables must be present for the further analysis:
- Gene expression values named as
expression - Systematic id (gene ids) named as
systematic_name
Delete observations that are missing or empty ("") in any of the two mandatory variables. How many rows did you remove? Save as cleaned_brauer
Representing the data
Tidying the data is a crucial step allowing easy handling and representing.
Plot the expression data of the LEU1 gene
Plot the data for the gene called in name LEU1 and draw a line for each nutrient showing the expression in function of the growth rate.
Plot the expression data of a biological process
For this, we don’t need to filter by single gene names as the raw data provides us some information on the biological process for each gene.
Extract all the genes in the leucine biosynthesis process (column BP) and plot the expression in function of the growth rate for each nutrient.
Perform a linear regression in top of the plots
Let’s play with the graph a little more. These trends look vaguely linear.
Add a linear regression with the appropriate ggplot2 function and carefully adjust the method argument.
Switch to another biological process
Once the dataset is tidy, it is very easy to switch to another biological process.
Instead of the “leucine biosynthesis”, plot the data corresponding to “sulfur metabolism”.
Tip
you can combine the facet headers using + in facet_wrap(). Adding the systematic name allows to get a name when the gene name is missing.