Fig. 4

Confusion matrices of interactions \(\hat{\theta }_{jj'}\) for the different methods, using the following simulation parameters: \(I = 5\), \(\sigma = 0.5\), \(n=100\). We can see from this example that MLGL and \(\rho \)-SICOMORE behave similarly, with very large genomic regions identified. SICOMORE tends to work with smaller genomic and metagenomic regions