Related Experiment Videos
Family pairwise search with embedded motif models
1Department of Computer Science, University of California, Santa Cruz, CA 95064 and NPACI/SDSC, MC 0505, 9500 Gilman Drive, Bldg 109, La Jolla, CA 92093-0505, USA. bgrundy@cse.ucsc.edu
Bioinformatics (Oxford, England)
|June 26, 1999
Summary
A new algorithm improves protein homology detection by combining Family Pairwise Search (FPS) with hybrid motif models. This method outperforms existing techniques, especially when limited training sequences are available.
Area of Science:
- Bioinformatics
- Computational Biology
- Genomics
Background:
- Statistical models like HMMs are effective for remote homology detection but require large training sets.
- Existing methods struggle with limited sequence data, performing worse than Family Pairwise Search (FPS).
Purpose of the Study:
- To develop a model-based algorithm that enhances homology detection performance using limited protein family data.
- To improve upon the Family Pairwise Search (FPS) algorithm by integrating hybrid motif models.
Main Methods:
- Developed a hybrid motif-based model by incorporating Cobbler-generated models into the Family Pairwise Search (FPS) framework.
- Evaluated the algorithm's performance across 73 diverse protein families.
Main Results:
- The cobbled FPS algorithm demonstrated superior homology detection compared to Cobbler or FPS alone.
- Performance improvements were consistent even when using the Smith-Waterman algorithm instead of BLAST.
Conclusions:
- The proposed model-based algorithm effectively enhances remote homology detection, particularly in low-data regimes.
- This approach offers a significant advancement for identifying protein family relationships with limited sequence information.