Comparison Between Profile-Profile Methods Based on Hidden Markov Models and Multiple Alignments

摘要

One of the feasible approaches for the detection of remote homologs is the profile-profile comparison that is based on the extraction of intrinsic features from the multiple alignments between the amino acid sequences in protein families by various methods. In this study, a profile-profile comparison method is designed to utilize the profiles in the match nodes of the hidden Markov model profiles with log average scoring scheme. The performance for the remote homolog detection of profiles derived from two different HMM programs is evaluated by comparing those of two HMM methods, and the another profile-profile method based on multiple alignments, COMPASS. As a result, the performance of our method is superior to the two HMM methods and comparable to COMPASS in the remote homolog detection, while the robustness is equivalent to each other. In addition, the correctly aligned length of our methods is slightly better than those of the other methods. Thus, the present results indicate the usability of the HMM profiles for the profile-profile comparison.