Related Experiment Videos
Cascaded multiple classifiers for secondary structure prediction
1Department of Computer Science, University of Wales, Ceredigion, United Kingdom. mho@aber.ac.uk
Protein Science : a Publication of the Protein Society
|July 13, 2000
Summary
A novel protein secondary structure classifier, using cascaded neural networks and linear discrimination, achieves 76.7% accuracy. This method highlights the effectiveness of local sequence information for predicting beta-strands.
Area of Science:
- Bioinformatics
- Computational Biology
- Structural Bioinformatics
Background:
- Accurate protein secondary structure prediction is crucial for understanding protein function and design.
- Existing methods face challenges with accuracy and dataset redundancy.
Purpose of the Study:
- To develop and evaluate a new, highly accurate protein secondary structure prediction classifier.
- To assess the role of local versus long-range interactions in secondary structure prediction.
Main Methods:
- A novel classifier combining neural networks and linear discrimination in a cascade architecture.
- Utilized a new, nonredundant dataset of 496 nonhomologous protein sequences.
- Employed a rigorous full Jack-knife cross-validation procedure for performance assessment.
Main Results:
- Achieved a prediction accuracy of 76.7% on the independent test set.
- Demonstrated high discrimination accuracy (up to 78%) for beta-strands using local sequence windows.
- Indicated that the importance of long-range interactions for beta-strand prediction may have been overestimated.
Conclusions:
- The proposed cascaded classifier offers improved accuracy in protein secondary structure prediction.
- Local sequence information and resampling techniques are highly effective for predicting beta-strands.
- This suggests a re-evaluation of the significance of long-range interactions in specific secondary structure element predictions.