Independent Study to prepare for workshop

Transcriptomics 3: Visualising

Emma Rand

17 December, 2024

Overview

In these slides we will:

  • Check where you are

  • learn some concepts used omics visualisation

    • Principle Component Analysis (PCA)
    • Volcano plots

  • Find out what packages to install before the workshop

Where should you be?

What we did in Transcriptomics 2: Statistical Analysis

  • carried out differential expression analysis

  • found genes not expressed at all, or expressed in one group only

  • Saved results files

Where should you be?

After the Transcriptomics 2: 👋 Statistical Analysis Workshop including:

🐸 Frog development

  • An RStudio Project called frogs-88H which contains:

    • data-raw: xlaevis_counts_S14.csv, xlaevis_counts_S20.csv, xlaevis_counts_S30.csv
    • data-processed: s30_filtered.csv, s20_filtered.csv
    • results: s30_fgf_only.csv (there were no control only genes in s30), s30_results.csv, and equivalent for S20
    • Two scripts: cont-fgf-s30.R, cont-fgf-s20.R

🎄 Arabidopisis

  • An RStudio Project called arab-88H which contains:

    • data-raw: arabidopsis-wild.csv, arabidopsis-spl7.csv
    • data-processed: wild_filtered.csv, spl7_filtered.csv
    • results: wild_suf_only.csv, wild-_def_only.csv, wild_results.csv, and equivalents for spl7
    • Two scripts: suff-def-wild.R, suff-def-spl7.R

💉 Leishmania

  • An RStudio Project called leish-88H which contains:

    • data-raw: leishmania-mex-ama.csv, leishmania-mex-pro.csv, leishmania-mex-meta.csv
    • data-processed: pro_meta_filtered.csv, pro_ama_filtered.csv
    • results: pro_meta_results.csv, pro_ama_results.csv
    • Two scripts: pro_meta.R, pro_ama.R

🐭 Stem cells

  • An RStudio Project called mice-88H which contains:

    • data-raw: surfaceome_hspc.csv, surfaceome_prog.csv, surfaceome_lthsc.csv
    • data-processed: hspc_prog.csv, hspc_lthsc.csv
    • results: hspc_prog_results.csv, hspc-lthsc_results.csv,
    • Two scripts: hspc-prog.R, hspc-lthsc.R

Additionally…

Files should be organised into folders. Code should well commented and easy to read. You should have curated your code to remove unnecessary commands that were useful to troubleshoot or understand objects in your environment but which are not needed for the final analysis.

If you are missing files, go through:

Go through:

Examine the results files

All results files

Remind yourself of the key columns in any of the results files:

  • normalised counts for each sample/cell
  • a log2 fold change
  • an unadjusted p-value
  • a p value adjusted for multiple testing (called FDR or padj)
  • a gene id
  • other information about each gene

🐸 , 🎄 , 💉 results files

  • baseMean is the mean of the normalised counts for the gene across all samples
  • lfcSE standard error of the fold change
  • stat is the test statistic (the Wald statistic)

🐭 Stem cells

  • Top is the rank of the gene ordered by the p-value (smallest first)
  • summary.logFC and logFC.hspc give the same value (in this case since comparing two cell types)

Plots

What is the purpose of a Transcriptomics plot?

  • In general, we plot data to help us summarise and understand it

  • This is especially import for transcriptomics data where we have a very large number of variables and often a large number of observations

  • We will look at two plots very commonly used in transcriptomics analysis: Principal Component Analysis (PCA) plot and Volcano Plots

Principal Component Analysis (PCA)

PCA

  • Principal Component Analysis is an unsupervised machine learning technique

  • Unsupervised methods1 are unsupervised in that they do not use/optimise to a particular output. The goal is to uncover structure. They do not test hypotheses

  • It is often used to visualise high dimensional data because it is a dimension reduction technique

PCA

  • Takes a large number of continuous variables (like gene expression) and reduces them to a smaller number of variables (called principal components) that explain most of the variation in the data

  • The principal components can be plotted to see how samples cluster together

PCA

  • To understand the logic of PCA, imagine we might plot the expression of one gene against that of another

Samples

Cells

This gives us some in insight in how the sample/cells cluster. But we have a lot of genes (even for the stem cells) to consider. How do we know if the pair we use is typical? How can we consider all the genes at once?

PCA

  • PCA is a solution for this - It takes a large number of continuous variables (like gene expression) and reduces them to a smaller number of “principal components” that explain most of the variation in the data.

Samples

Cells

PCA

We have done PCA after differential expression, but often PCA might is one of the first exploratory steps because it gives you an idea whether you expect general patterns in gene expression that distinguish groups.

Volcano plots

Volcano plots

  • Volcano plots often used to visualise the results of differential expression analysis

  • They are just a scatter of the adjusted p value against the fold change….

  • almost - in fact we plot the negative log of the adjusted p value against the log fold change

Volcano plots

  • This is because small probabilities are important, large ones are not so the axis is counter intuitive because small p-values (i.e., significant values) are at the bottom of the axis)

  • And since p-values range from 1 to very tiny the important points are all squashed at the bottom of the axis

Volcano plot padj against fold change

Volcano plots

  • By plotting the negative log of the adjusted p-value the values are spread out, and the most significant are at the top of the axis

Volcano plot -log(adjusted p) against fold change

Visualisations

  • Should be done on normalised data so meaningful comparisons can be made

  • The 🐭 stem cell data were already log2normalised

  • The other datasets were normalised by the DE method and we saved the values to the results files. We will log transform them in the workshop

Packages

This package is on the University computers which you can access on campus or remotely using the VDS

If you want to use your own machine you will need to install the package.

Install ggrepel from CRAN in the the normal way:

install.packages("ggrepel")

This package allows you to label points on a plot without them overlapping.

Workshops

Workshops

  • Transcriptomics 1: Hello data Getting to know the data. Checking the distributions of values overall, across rows and columns to check things are as we expect and detect rows/columns that need to be removed

  • Transcriptomics 2: Statistical Analysis. Identifying which genes are differentially expressed between treatments. This is the main analysis step. We will use different methods for bulk and single cell data.

  • Transcriptomics 3: Visualising. Principal Component Analysis (PCA) volcano plots to visualise the results of the

References

Rand, Emma. 2021. Data Science Strand of BIO00058M. https://doi.org/10.5281/zenodo.5527705.