Automatic Classification of Protein Structures
Author Information
Author(s): Sam Vichetra, Tai Chin-Hsien, Garnier Jean, Gibrat Jean-Francois, Lee Byungkook, Munson Peter J
Primary Institution: National Institutes of Health
Hypothesis
The study investigates how different clustering methods affect the classification of protein structural domains based on pairwise similarity scores.
Conclusion
Different clustering procedures give rise to different levels of agreement between automatic and manual protein classifications.
Supporting Evidence
- Ward's method dendrograms led to partitions closest to the SCOP fold classification.
- Two strategies optimizing similarity to SCOP achieved an average of 72% true positives rate at a 1% false positive rate.
- Visual examination confirmed that global structural similarity is not the only criterion used in SCOP classification.
Takeaway
The study looks at how to automatically group proteins based on their shapes, but finds that it's hard to match human classifications perfectly.
Methodology
The study used hierarchical clustering methods on pairwise similarity scores from three different structure comparison methods (VAST, SHEBA, DALI) to classify protein domains.
Limitations
The study acknowledges that achieving full agreement between automatic and manual classifications requires additional information.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website