Optimizing Amino Acid Substitution Matrices
Author Information
Author(s): Saigo Hiroto, Vert Jean-Philippe, Akutsu Tatsuya
Primary Institution: Bioinformatics Center, Institute for Chemical Research, Kyoto University
Hypothesis
The local alignment kernel can be optimized for better performance in detecting remote homologs compared to traditional methods.
Conclusion
The optimization procedure leads to useful substitution matrices that improve homology detection performance.
Supporting Evidence
- The local alignment kernel outperformed the Smith-Waterman algorithm in detecting remote homologs.
- Optimized matrices showed significant improvements in confidence values for homology detection.
- The study utilized a dataset of protein sequences with less than 20% identity for training and testing.
Takeaway
This study shows how to make better tools for comparing proteins by improving the way we score their similarities.
Methodology
The study used gradient descent to optimize amino acid substitution matrices based on a local alignment kernel.
Limitations
The optimization may not generalize well to all types of sequence comparisons.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website