Understanding Correlation Analysis

When to use correlation analysis

Correlation analysis is useful when your question is about coordinated behavior rather than group differences. In CytokineProfile Shiny, it helps answer questions such as:

Which cytokines track most strongly with a target marker?
Do those relationships look linear or only monotonic?
Do correlations change between two biological groups?

When not to use correlation analysis

Correlation analysis is usually not the best first choice when:

your main goal is comparing group averages rather than associations
the response of interest is categorical rather than numeric
you need prediction or classification rather than association screening

What the app is showing

The correlation workflow centers on one numeric response variable and compares it with all other numeric variables in the dataset.

Depending on the settings, the app can return:

overall correlation tables
Pearson and Spearman correlation views
optional subgroup-specific heatmaps
a heatmap-style plot of the correlation structure

Which app arguments matter most

The most important settings are:

Response Variable: the numeric measurement you want to compare against the rest of the dataset.
Stratification Variable: the optional grouping column used to split the correlations into subgroups.
Per-group Heatmaps: whether the app should render separate subgroup-specific heatmaps.

Choosing Pearson versus Spearman

The app shows both Pearson and Spearman results, so it helps to read them together:

Pearson correlation is best when the relationship is approximately linear and sensitive to magnitude.
Spearman correlation is best when the relationship is monotonic but not necessarily linear, or when ranking is more reliable than raw scale.

For many biological datasets with skewness or outliers, Spearman is often a safer first look.

How to interpret the results

A good reading order is:

Identify the strongest positive and negative correlations in the overall table.
Compare the Pearson and Spearman views to see whether the relationship looks consistently strong across methods.
If a Stratification Variable is selected, check whether those same relationships are consistent across subgroups.

Interpretation tips:

A large absolute correlation coefficient suggests a stronger association.
A small p-value suggests the observed association is less consistent with no correlation.
Group differences in correlation can be biologically meaningful even when the overall correlation looks modest.

The subgroup-specific view is especially valuable because it can reveal relationships that are masked when all samples are pooled together.

Common cautions

Important cautions include:

Correlation does not imply causation.
Strong correlations can be driven by outliers or narrow sample ranges.
Pooling heterogeneous groups can create misleading correlations.
Multiple testing still matters when screening one target against many variables.

How to reproduce the result in the app

Choose a numeric target variable with biological interest.
Choose Correlation Plots.
Set Response Variable.
Add a Stratification Variable and turn on Per-group Heatmaps if you want to compare biological subgroups.
Use the table and the plot together to decide which associations deserve follow-up.