Improving Protein Structure Prediction with Pairwise Covariance
Author Information
Author(s): Christopher Bystroff, Bobbie-Jo Webb-Robertson
Primary Institution: Rensselaer Polytechnic Institute
Hypothesis
Can pairwise covariant sequence models improve the prediction of protein local structures compared to profile-based models?
Conclusion
Pairwise covariance improves the prediction of certain loop motifs but does not significantly enhance the prediction of alpha helices or beta strands.
Supporting Evidence
- Pairwise covariances were shown to be statistically robust in cross-validation tests.
- The study found that reducing the amino acid alphabet to nine classes was critical for statistical significance.
- Improvements in local structure prediction were observed when covariance was added to the profile model.
Takeaway
This study looks at how well we can guess the shapes of proteins based on their building blocks, and it finds that looking at pairs of building blocks helps with some shapes but not all.
Methodology
The study used a modified I-sites motif library with pairwise covariance metrics to predict local protein structures and evaluated the predictions on a test set of proteins.
Limitations
The study found that while pairwise covariance improved predictions for non-canonical structures, it did not significantly enhance predictions for canonical secondary structures.
Statistical Information
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website