Abstract
Abstract After the gnome sequencing project of human and many other species has completed, a lot of protein sequences are also rapidly discovered and accumulated. In the post genome era, it is an urgent task to identify the key regions in a protein(enzyme) sequence that have significant effects on protein(enzyme) function. One of the straightforward ways is to analyze protein structures by experiments and then to understand functional mechanisms based on the structural information. But the analysis of protein structure is very costly with both time and expenses; identifying enzyme functional residues based on amino acid sequence is often demanded in many biological studies especially when related structure information is not available. In some cases of protein superfamilies, the functional residues could hardly be detected by multiple sequence alignment or evolutionary strategies when phylogenetic relationships do not parallel to their protein functions. In this study, a feature amplified voting algorithm (FAVA) is developed to solve such a problem. Given a target sequence and several related protein sequences (usually they belong to the same protein superfamily), we divide related sequences into two sets according to their functional properties and perform sequence alignments with target sequence. After alignment phase, functional residues of target protein can then be extracted by following voting analysis. The main method of sequence alignment is simultaneous three sequences alignment; besides, the FAVA concept is also applied to several well-known multiple sequence alignment methods that has been widely used. Finally the results from different methods are compared and analyzed. Amidohydrolase superfamily is used as a case study because it contains divergent enzymes and proteins and provides an interesting case for developing such method. FAVA is used to identify critical residues of mammalian imidase, a member of amidohydrolase superfamily. In 519 amino acids of rat imidase sequence, we predict 10 candidate residues, five residues among them are related to metal binding. In this study, we have verified that the voting concept can also be combined with other multiple sequence alignment method and improve their performance on functional residues prediction.