Family specific protein sequence scoring matrices and applications.

摘要

All the alignment methods require an alignment algorithm and scoring scheme that include substitution scores and gap penalties. It is sometimes suggested that the proper scoring matrix is the most critical technical element in a successful protein database search. Any improvement would benefit to all the processes mentioned above. This dissertation investigates a new way to construct family specific scoring matrix and constructs scoring matrices for different protein structures then tests their performance in terms of sequence alignment and database search.; We extended Dayhoff's work and estimated amino acid substitution model for individual protein family from alignments of sequences with varying degree of divergence using a maximum likelihood method. Different protein families showed some unique mutation patterns. The performance test showed that family specific matrix performed significantly better than other matrices in the detection of remote homologues. However, the family specific matrix did not improve the alignment quality significantly. The cluster analysis of enzymes in two different pathways showed that although the enzymes in the same pathway were likely to be grouped closer together, the separation was not perfect. The structure similarity may be the key reason for such separation.; We derived the scoring matrices for different structures as defined in OATH database since different structures should have different substitution patterns. The cluster analysis of those matrices constructed at architecture level showed that the relationship between these matrices were in general consistent with the OATH classification. Structure-based matrix significantly improved the sequence alignment quality, this was specially true for the matrix specific to a particular structure architecture. Structure-based matrix also performed better than BIOSUM matrix in the fold recognition test; for the TIM barrel fold and Rossmann fold, the fold specific matrices were the best performers. Finally, we showed that such structure based matrices can be useful in the genome annotation.; The matrices derived in this study can be used in the other database search program such as PSI-BLAST to improve the accuracy. The cluster analysis of mutation matrices contributes to our understanding of metabolic evolution and provides impetus for further research in the pathway evolution.