Should I use z-score in marker expression heatmaps?

[Algebio]

Hi all

I am analysing imc data using mainly CATALYST (from Helena Crowell). This package doesn't use z-score until the last heatmap to study differential expression. I am having difficulties to understand when to use the different scaling and transformation options. For instance, in CATALYST I'm trying to understand why the first heatmap uses the scaled median expression, the second the median scaled, the third one the normalize frequency, same for DA heatmap, and the DS heatmap uses the z-normalized expression.

I have special interest for the z-normalized expression because I have been asked to produce every heatmap using z-score. Firstly; How could I do it? I have tried to modify the different heatmaps in CATALYST without success, I have tried scater::plotHeatmap, I have tried singlecellTK::plotSCEHeatmap but it produces a huge vector that my laptop can't handle. Do you have any suggestion to heatmap a (fairly big) singlecellexperiment using z-score?

And secondly; Is there any reason why I shouldn't do it in every heatmap? I guess there is when they don't do it with CATALYST but I need this information if I'm not going to z-score every heatmap. On the other hand, a recent imc paper by A. Rendeiro uses z-score in the heatmap used to identify cell populations.

I am confused and any help will be very welcome.

Regards
juan

[tomash]

Hi Juan,

Without knowing exactly what's trying to be visualised it is a little hard to say, but different scaling might be used on different heatmap when trying to show different things -- for example if one is showing expression levels of markers, then a simple min/max signal intensity scaling would suffice. If trying to show fold change for populations across experimental groups, then perhaps z-score might be used etc. It's a vague answer I know, but I guess the point is that that different scaling might be appropriate for different purposes.

In terms of help with that specific issue creating that heatmap in CATALYST, you can always submit an 'issue' on their Github: https://github.com/HelenaLC/CATALYST/issues. That way the authors/maintainers can comment directly.

If you want some more flexibility in the way you make heatmaps and how you scale the data, you can also check out our package Spectre (https://immunedynamics.io/spectre/) -- you basically just manage the data as a large table (cells vs markers) and can apply whatever scaling you like, include z-score. Then you can make whatever type of heatmap you like. See a quick tutorial here: https://wiki.centenary.org.au/display/S ... g+heatmaps.

[ta3limynet]

The decision to use z-scores in marker expression heatmaps depends on your specific goals and the characteristics of your data. Z-scores (standard scores) are often used in heatmaps to standardize data and make it comparable across different markers or conditions. Here are some considerations:

Advantages of using z-scores:

Normalization: Z-scores standardize data by subtracting the mean and dividing by the standard deviation. This can be particularly useful when markers have different scales or variances. Normalizing data allows for better comparison between markers.

Visual Comparison: Z-scores can help highlight patterns of upregulation or downregulation across different markers. This can make it easier to visually identify trends and patterns in the data.

Interpretability: Z-scores represent the number of standard deviations a data point is from the mean. This can provide a more interpretable scale for comparing marker expression levels.

Considerations:

Data Distribution: Before applying z-scores, it's important to assess the distribution of your data. If the data is not normally distributed or if there are outliers, z-scores might not be the most appropriate choice. In such cases, alternative normalization methods or transformation techniques may be considered.

Biological Interpretation: The choice of normalization method should also be guided by the biological context. Some researchers prefer not to use z-scores if the raw expression values have clear biological meaning, and normalization might obscure important biological trends.

Context of Analysis: Consider the context in which you are conducting your analysis. If the goal is to compare marker expression across different conditions or samples, z-scores may be beneficial. However, if the goal is to examine the absolute expression levels of markers, you might prefer not to use z-scores.

Software and Visualization Tools: Check the capabilities of the software or tools you are using for heatmap generation. Some tools automatically apply z-score normalization, while others may offer alternative normalization options.

In summary, using z-scores in marker expression heatmaps can be a valuable approach, especially for comparing expression patterns across different markers or conditions. However, it's essential to carefully consider the characteristics of your data and the goals of your analysis before deciding on the most appropriate normalization method.