Clustering and artificial neural networks: classification of variable lengths of Helminth antigens in set of domains

Abstract

A new scheme for representing proteins of different lengths in number of amino acids that can be presented to a fixed number of inputs Artificial Neural Networks (ANNs) speel-out classification is described. K-Means's clustering of the new vectors with subsequent classification was then possible with the dimension reduction technique Principal Component Analysis applied previously. The new representation scheme was applied to a set of 112 antigens sequences from several parasitic helminths, selected in the National Center for Biotechnology Information and classified into fourth different groups. This bioinformatic tool permitted the establishment of a good correlation with domains that are already well characterized, regardless of the differences between the sequences that were confirmed by the PFAM database. Additionally, sequences were grouped according to their similarity, confirmed by hierarchical clustering using ClustalW.

Vector Space Information Retrieval Techniques for Bioinformatics Data Mining

“…Finally, for the sake of clarity, we restrict our attention to sequences having the same lengths. The extension of these results to variable length sequences is the subject of current research based upon existing methodologies cited in the literature Couto et al (2007); T. Rodrigues (2004). The histogram in Figure 4 illustrates the number of BLOCKS families as function of the number of sequences contained within in each family; however, observe that this representative sample has been restricted to those families containing sequences of equal length (in this case L = 30).…”