So I've recently been involved in an RNAseq project and my investigator is interested in comparing RNA abundances in organoids to corresponding tissue samples taken directly from the organism (a mouse). There were three experimental groups for both tissue types as well to test the effects of these treatments on both the organism and the organoid. We are primarily interested in arguing that an organoid is an appropriate model for the whole mouse. We would like to demonstrate that there are genes that exhibit similar differential abundance patterns in both the organism and the organoid for each of the three treatments.
As expected, the transcriptome of an organoid is very different from the corresponding original host tissue, so most heatmaps are going to illustrate differences between organoid and tissue rather that differences of responses to the experimental treatments. In an ideal world, samples from treatment A would look similar (on a heatmap) regardless of whether they came from a tissue sample or an organoid. Alternatively, if the samples were to be divided into two heatmaps, one for organoid samples and one for tissue samples, the pattern of color contrasts between treatments A and B would be similar on both heatmaps.
I have been playing around with different gene sets and cutoffs and I am still unsure where to go next. Are there any sorting techniques similar to unsupervised clustering that are designed to emphasize differences along one dimension and group according to similarity along the other? Perhaps there is a specific technique to select a subset of genes that will generate a heatmap in which treatment contrast is emphasized over organoid contrast. Has anyone run into similar problems? Are there any established protocols or tools that you have used to generate heatmaps according to unique constraints? Are there other bioinformaticians that have run into similar problems?
Please feel free to share your experiences, suggestions, and opinions
As you rightly said, expression levels are by default different between the organoids and the primary tissues.
Assuming you are using R, I may try the flow below;
1. Assume the primary tissue is the ground truth.
2. Perform differential expression (DE) analysis on the primary tissue treatment group and extract the differentially expressed genes.
3. Perform DE analysis on the organoid groups and extract the corresponding expression of those primary tissue DE genes.
Note: You may need to also include genes differentially expressed in just the organoids to show that future experiments will not retrieve unimportant genes.
4. Using the ComplexHeat package on R, create a heat map list (having the primary tissue matrix as the main heat map and the organoid matrix as the secondary). You can use any distance and linkage combination from the hclust function.
5. Annotate the secondary organoid heat map with the p-values from its DE analysis.
You can also scale the colour mapping of the two heat maps separately as well as add other annotations.
Ideally, you should have something like the first attached image. This should convey the information.
Alternatively, you can use an UpSet plot function from the ComplexHeatmap package.
1. Perform DE analysis and extraction as before.
2. For the selected genes, calculate relevant ratios.
3. Create an Upset plot and decorate with other important information.
Ideally, you get something like the second attached image to convey the association between the tissue groups.
Thanks Chinedu, I'll give ComplexHeat a try, sounds like that package has a lot of added functionality that I would be interested in checking out. I'll also give weighted clustering a try.
David Johanson you are welcome. Weighted clustering is also an option.
You may want to try to select a sensible gene list to start with. A gene list selected after Simple Effect Contrasts or Interaction Contrasts will be a good starting point in an experiment with multiple factors. See attached slide for definitions and a reference to read more.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics