R: heatmap

2021-07-09

The representation of associations between two classes of features becomes complex when the number in each class increases and this occurs for instance when investigating:

• genes to clinical parameters.
• mRNA to miRNA.
• subjects to lifestyles.

Heatmap is a regular proposition, but can be confusing, and it should be carefully designed so as to help identifying the most relevant associations between individual features or group of features. In the example below:

• The association between a row feature and a column feature is quantified at the intersection.
• The intensity of the shade corresponds to the absolute value of the correlation.
• We should always understand the rational for row or column ordering: in that case, an unsupervised hierarchical classification was applied; within rows (or columns) to close features (distance derived from correlation) will be found grouped together. The information is supported by the addition of the dendrogram in the margins.
• The numbers are the correlations; number by themselves may not be of main interest so the first visible information is actually the colour of the numbers corresponding to the discrimination of significance (yellow, p.value < 0.5; green otherwise). Still, looking closer at the matrix give access to the actual correlation value if required (e.g. for reporting).

From that, we understand that the rows and columns features under the magnifier glass constitute two groups which association support their identification as good candidates to push the analysis further.