02/01/1998 09:00 AM
Computer Science
A new HMM-based method (target98) for finding remote homologs of protein sequences is evaluated using four test sets. Three of the test sets are fold-recognition tests, where the correct answers are determined by structural similarity. The fourth uses a curated database. The method is compared against WU-BLASTP and DOUBLE-BLAST, a two-step method similar to ISS, but using BLAST instead of FASTA. The HMM-based method had the fewest errors in all tests--- dramatically so for the fold-recognition tests. One key to the performance of the HMM method is a new score-normalization technique which compares the score to the score of the reversed sequence, rather than to a null model.
