A new seed selection algorithm that maximizes local structural similarity in proteins.

The PHI-BLAST algorithm for protein sequence alignment takes a query sequence and searches a protein database for a small seed or region of high similarity and extends this alignment to produce the total alignment for sequences. Clearly, the success of this approach depends on the quality of the seeds. We propose an algorithm that maximizes the likelihood of seeds sharing the same local structure in both the query and known sequences. This was tested on the 2290 protein sequences in the PISCES database. Our new algorithm results in an effective a priori estimate of seed structural quality.