FastCluster: a graph theory based algorithm for removing redundant sequences

Abstract

In many cases, biological sequence databases contain redundant sequences that make it difficult to achieve reliable statistical analysis. Removing the redundant sequences to find all the real protein families and their representatives from a large sequences dataset is quite important in bioinformatics. The problem of removing redundant protein sequences can be modeled as finding the maximum independent set from a graph, which is a NP problem in Mathematics. This paper presents a novel program named FastCluster on the basis of mathematical graph theory. The algorithm makes an improvement to Hobohm and Sanders algorithm to generate non-redundant protein sequence sets. FastCluster uses BLAST to determine the similarity between two sequences in order to get better sequence similarity. The algorithms performance is compared with Hobohm and Sanders algorithm and it shows that Fast- Cluster can produce a reasonable non-redundant protein set and have a similarity cut-off from 0.0 to 1.0. The proposed algorithm shows its superiority in generating a larger maximal non-redundant (independent) protein set which is closer to the real result (the maximum independent set of a graph) that means all the protein families are clustered. This makes Fast- Cluster a valuable tool for removing redundant protein sequences.