Related Experiment Videos
Profile-based direct kernels for remote homology detection and fold recognition.
Huzefa Rangwala1, George Karypis
1Department of Computer Science and Engineering, University of Minnesota Minneapolis, MN 55455, USA.
Bioinformatics (Oxford, England)
|September 29, 2005
Summary
New kernel functions significantly improve protein remote homology detection using support vector machines (SVMs). These novel kernels outperform existing methods, even without sequence profiles, advancing computational biology.
Area of Science:
- Computational Biology
- Bioinformatics
- Machine Learning
Background:
- Protein remote homology detection is crucial in computational biology.
- Support vector machines (SVMs) are effective for this task.
- Method performance relies on protein sequence modeling and kernel function computation.
Purpose of the Study:
- Introduce novel kernel functions for enhanced protein remote homology detection.
- Improve upon existing SVM-based methods for sequence similarity analysis.
Main Methods:
- Developed two classes of kernel functions combining sequence profiles with similarity measures.
- Employed effective profile-to-profile scoring schemes for protein pair similarity.
- Utilized SVMs for classification tasks.
Main Results:
- The new kernels achieved substantially better results than state-of-the-art SVM methods.
- These kernels outperformed existing non-profile-based schemes, even without profiles.
- Demonstrated effectiveness in remote homology detection and fold recognition.
Conclusions:
- The proposed kernel functions represent a significant advancement in SVM-based protein analysis.
- These kernels offer improved accuracy for remote homology detection and fold recognition.
- The methods provide a more powerful tool for computational biology research.