ke.ca

ooke.ca We are interested in the problem of grouping families of non-alignable protein sequences, such as circular-permutation, multi-domain and tandem-repeat proteins, into clusters (classes) of related biological functions. For such sequences, whose numbers are constantly growing, the commonly used alignment-dependent approaches fail to yield biologically plausible results. To the best of our knowledge, no automatic process yet exists to carry out clustering on these proteins. Biologists often use more complex manual approaches based on secondary and tertiary structures, which require considerably more resources and time. In this paper, we develop a new similarity measure SMS, applied directly on non-aligned sequences. It allows us to develop a new and original alignment-free algorithm, named CLUSS, for clustering protein families based on a spectral decomposition approach inspired by the latent semantic analysis (LSA) widely used in information retrieval. CLUSS, utilized jointly with SMS, is effective on both alignable and non-alignable protein sequences. To show this, we have extensively tested our algorithm on different benchmark protein databases and families; we have also compared its performance with many alignment-dependent mainstream algorithms. The source code, the application server, and all experimental results are available at CLUSS web sit