Algorithm 1: Heatmap based on mutual conditional entropy

Input: $N\times K$ data matrix ${\mathcal{M}}_0=\left[X_k\left[i\right]\right]$. Output: The heatmap $M_{ce}^{\mathcal{T}}$. For each pair k and $k^{\prime }$, compute $\mathcal{E}\left[X_k^c\right|X_{k^{\prime }}^c]=\frac{H\left[X_k^c\right|X_{k^{\prime }}^c]}{H\left[X_k^c\right]}$. The $K\times K$ mutual conditional entropy matrix is defined as $M_{ce}=[\mathcal{E}<k,k^{\prime }>]$, where $\mathcal{E}<k,k^{\prime }>=[\mathcal{E}\left[X_k^c\right|X_{k^{\prime }}^c]+\mathcal{E}\left[X_{k^{\prime }}^c\right|X_k^c]]/2$. Apply Hierarchical clustering algorithm on $M_{ce}$ to build a binary HC-tree ${\mathcal{T}}_{mce}$. Permute the rows and columns of $M_{ce}$ according to the arrangement of tree-leaves of ${\mathcal{T}}_{mce}$ to have $M_{ce}^{\mathcal{T}}$.