Think about how to visualize performance across comparisons

[sgosline]

How do we want to plot the correlations or mutual information?

[sgosline]

Discussed this on 12/31. We want to evaluate the stability of both algorithm and signature matrix in three different dimensions.

First: how stable is an algorithm across the mRNA datasets; This is a pairwise correlation matrix that shows how similar algorithms are (one correlation matrix/heatmap for each signature matrix)

Second: how stable is an algorithm between mRNA and protein. This can be answered in two ways: 1- compute correlation across all patients for full datasets 2- compute correlation across all patients for each cell type and plot in barplot (which cell types agree most) 3- compute correlation across cell types by patient - use boxplot for correlation values by cancer type (which cancer types agree most)

Third: how stable is a signature matrix between mRNA and protein. For this we repeat the same comparisons for each algorithm-- 1- compute correlation across patients (one for each cell type) and plot in barplot (which cell types agree most) 2- compute correlation across cell types by patient - use boxplot for correlation values by cancer type (which cancer types agree most)

[LifeWorks]

Implemented the heatmap with all the correlations across patients and cell types. Comparisons: 1) algorithms + signature vs patients for each cancer type; 2) algorithms + cancer type vs cell types for each signature matrix.

I think these two sets of heatmaps captures all three sections of comparison. If not, maybe we should discuss about it.

@sgosline Please check the results folder.