Table 1 Biclustering search strategies as defined by Madeira and Oliveira [7]
Category | General characteristics | Examples |
|---|---|---|
Greedy | Biclusters are generated by adding or removing columns to a initial random bicluster in order to improve some gain function. The final objective is for the algorithm to find a global minimun solution after some iterations. Despite making wrong decisions, and loosing good biclusters due to beeing stuck in local minima, they have the potential of being fast algorithms | |
Distribution parameter identification | Assume some statistical model behind the data, and then apply some iterative procedure in order to obtain its parameters by minimizing some criterion | |
Divide and conquer | Divide the original data matrix into smaller instances. With the potential of being very fast, they could fail to find good biclusters, splitted before identified | Bimax [30] |
Exhaustive | Based on the premise that finding the best biclusters can only be done by using an exhaustive enumeration of all possible biclusters in the data matrix. Despite being able to find the bests biclusters they do it by imposing restrictions to the biclusters size (since these algorithms are typically very slow) |