Related Experiment Videos
A comprehensive analysis of 40 blind protein structure predictions
Ram Samudrala1, Michael Levitt
1Department of Microbiology, University of Washington, School of Medicine, Seattle, WA 98195, USA. ram@compbio.washington.edu
BMC Structural Biology
|August 2, 2002
Summary
This study evaluated 40 protein structure predictions from CASP4, finding model accuracy correlates with sequence identity. These findings inform future genome-wide protein structure modeling efforts.
Area of Science:
- Computational Biology
- Structural Bioinformatics
- Genomics
Background:
- Analysis of 40 blind predictions from the Critical Assessment of Protein Structure Methods (CASP4).
- Utilized comparative modeling, fold recognition, and ab initio methodologies for protein structure prediction.
- Predictions covered targets with sequence identities from 10% to 50% and those with no detectable sequence relationships.
Purpose of the Study:
- To assess the performance of protein structure prediction methods.
- To identify areas of success and areas needing improvement in protein structure prediction.
- To understand the implications for modeling entire proteomes.
Main Methods:
- Comparative modeling and fold recognition for 29 targets with detectable sequence similarity.
- Ab initio modeling for 11 targets lacking detectable sequence similarity.
- Evaluation of model accuracy using C-alpha root mean square deviation (RMSD).
Main Results:
- Generated 23 models with C-alpha RMSD ranging from 1.0 to 6.0 Å.
- Model accuracy showed a near-linear relationship with sequence identity.
- Achieved 4.0 Å C-alpha RMSD for residues 1-80 in T110/rbfa using ab initio methods.
Conclusions:
- Performance analysis across four CASP experiments reveals strengths and weaknesses in protein structure prediction.
- Established methods show promise for modeling tractable proteins across a genome.
- Further refinement of prediction techniques is necessary for broader applicability.