Skip to main content

Table 1 MI3 and control methods evaluated and compared using the synthetic data.

From: Learning transcriptional regulatory networks from high throughput gene expression data using continuous three-way mutual information

Method

Metric

Description

Performance Rank

   

Synth

Real #

MI3

2*I(T;R1, R2)-I(T;R1)-I(T; R2) = I(T; R1|R2)+ I(T; R2|R1)

The sum of Correlative and Coordinative Criteria, which equals to the conditional mutual information between the target gene and the each regulator given the other regulator

1

1

dMI3

2*I(T;R1, R2)-I(T;R1)-I(T; R2)

Discrete version of MI3, control score to show the strength of continuous mutual information

3

2

Bayesian network (BN)

logP(T | R1, R2)†

Log conditional probability, control score which maximize correlation of the parent set to the target, while ignores the interaction between R1 and R2

2

3

Two-way MI (MI2)

I(T;R1)+I(T;R2)

Control two-way mutual information score to show the strength of three-way metric

4

4

  1. Note that each control method compares to and validates MI3 in one of the three major aspects described in Introduction: data discretization (dMI3); pairwise testing (MI2); emphasis on correlation over causality (BN). All scores are calculated based on continuous nonparametric probability density estimation, except dMI3 based on discretization using 5 bins of equal size.
  2. † In this paper, log conditional probability and BN are used interchangeably.
  3. # Performance rank for real data experiment is based on qualitative comparison.