Liu2006

CREATED: 200612290326 LINK: url:~/Modules/Literature/Liu2006.pdf WWW: http://doi.ieeecomputersociety.org/10.1109/ICTAI.2006.74

** Background spatial temporal regulation

gene network (G -> G) as a simplification of cellular network (G -> P -> M, M -> P, M -> G, P -> G)

** Learning gene network

• pairwise correlation
• boolean models
• linear models
• nonlinear models
• differential models
• bayesian models
• causal influence
• data of a stochastic nature
• small datasets

** Challenges

• large number of dimensions
• noise
• randomness

** Learning strategy for Bayesian models *** candidate parent selection

• typically using pairwise correlation
• problems: weak parents, gene/protein complexes, variable time delay

*** determine actual parents using heuristics *** CD algorithm use CRE to find weak parents in candidate parent selection

• conditional dependence: A -> C <- B, if C is known, A and B are dependent
• conditional relative entropy (CRE)

find complexes in candidate parent set of z

• complexes as semi clique
• each semi clique of size 4 contains a clique of size 3 ** first find all 3 cliques by brute force ** use heuristics to extend the 3 cliques to semi cliques

use complexes and individual genes to check for parents