Mutual information (MI) represents the interdependence of two discrete random variables and is analogous to covariation in continuous data. The intersection of entropy space of two random variables bound MI and quantifies the reduction in uncertainty of one variable given the knowledge of a second variable. However, MI must be normalized by a leveling ratio to account for the background distribution arising from the stochastic pairing of independent, random sites. Martin et al. (2005) found that the background MI, particularly from phylogenetic covariation, has a contributable effect for multiple sequence alignments (MSAs) with less than 125 to 150 sequences.
NMI provides several methods for normalizing mutual information given the individual and joint entropies.
1 |
Hx |
Marginal entropy for a discrete random variable (x) |
Hy |
Marginal entropy for a discrete random variable (y) |
Hxy |
Joint entropy for a discrete random variables (x and y) |
type |
method of normalization. Default is "NULL" and the Mutual Information is calculated as MI = Hx+Hy-Hxy. Other methods include "marginal", "joint", "min.marginal", "max.marginal", "min.conditional", "max.conditional". See details below. |
If any denominator is zero, MI=0. Otherwise
Methods of Normalization:
marginal MI = 2*( Hx + Hy - Hxy ) / ( Hx + Hy ) joint MI = 2*( Hx + Hy - Hxy ) / ( Hxy ) min.marginal MI = ( Hx + Hy - Hxy ) / min(Hx,Hy) max.marginal MI = ( Hx + Hy - Hxy ) / max(Hx,Hy) min.conditional MI = ( Hx + Hy - Hxy ) / min(Hx.y,Hy.x) max.conditional MI = ( Hx + Hy - Hxy ) / max(Hx.y,Hy.x)
normalized mutual information value
Lisa McFerrin
Martin, L.C., G. B. Gloor, et al. (2005). Using information theory to search for co-evolving residues in proteins. Bioinformatics. 21, 4116-24.
MolecularEntropy
, MolecularMI
,
Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.
All documentation is copyright its authors; we didn't write any of that.