| |
| Bi-clustering is a hard problem in most research fields, especially in the study of gene expression patterns. Except the shifting patterns, along the broadening observations in the study of gene regulatory pathways, scaling patterns or even more general linear patterns are thought to be at least equal importance to investigate. Comparing with MSR only can identify shifting patterns, MMSE is a unified measurement to formal such three types of gene expression patterns. However, MMSE was originally used in node-add bi-clustering framework so that it’s not efficient enough. And in another way, most previous bi-clustering methods only output several first arriving bi-clusters to avoid redundancy in obtained bi-clusters, which can’t completely reveal the distribution of whole potential patterns.
So this paper proposes the error-bounded linear gene expression patterns based on MMSE and an effective bi-clustering method (Error-bounded Bi-clustering abbreviate as EB) to enumerate these refined patterns. A widely experiments on 54 controlled synthesized data-sets and 3 yeast cell cycle data-sets strongly support that EB is well matched in strength of currently state-of-the-art clustering/bi-clustering methods according to their biological P-value evaluations; and EB also has the most significant background clusters’ recovery ability than many other bi-clustering methods. |
|
| Keywords:gene expression;error-bounded linear pattern;bi-clustering |
|
