Related Experiment Videos
Evaluation and improvement of multiple sequence methods for protein secondary structure prediction
1Laboratory of Molecular Biophysics, Oxford, United Kingdom.
Proteins
|March 19, 1999
Summary
A new consensus method improves protein secondary structure prediction accuracy to 72.9% Q3, outperforming single algorithms. This study also introduces new datasets and highlights the importance of consistent state reduction for accurate algorithm comparison.
Area of Science:
- Structural bioinformatics
- Computational biology
- Protein structure prediction
Background:
- Accurate prediction of protein secondary structure is crucial for understanding protein function and evolution.
- Existing prediction algorithms vary in performance, and direct comparison is often hindered by differing evaluation methods.
Purpose of the Study:
- To evaluate and improve protein secondary structure prediction algorithms.
- To establish standardized datasets and methods for cross-validation.
- To develop a consensus prediction approach for enhanced accuracy.
Main Methods:
- Development of a new dataset comprising 396 protein domains.
- Evaluation of DSC, PHD, NNSSP, and PREDATOR algorithms.
- Creation of a consensus prediction method using automatically generated multiple sequence alignments.
- Derivation of new cross-validation datasets (CB513 and CB251).
Main Results:
- The consensus method achieved an average Q3 prediction accuracy of 72.9%, a 1% improvement over the best single method (PHD).
- Segment Overlap Accuracy (SOV) for the consensus method reached 75.4%.
- Analysis revealed over 3% variation in apparent accuracy due to different 8- to 3-state reduction methods for secondary structure definitions.
Conclusions:
- A consensus approach significantly enhances protein secondary structure prediction accuracy.
- Standardized datasets and reduction methods are essential for reliable algorithm benchmarking.
- A web service is available for automated protein secondary structure prediction.