Local Protein Structure Prediction Using Discriminative Models
Author Information
Author(s): Sander Oliver, Sommer Ingolf, Lengauer Thomas
Primary Institution: Max-Planck-Institute for Informatics
Hypothesis
Given a local sequence fragment, how much can we learn about the local structure it adopts?
Conclusion
The method provides probability estimates for local protein structure candidates, which can enhance the accuracy of fold recognition and new fold prediction methods.
Supporting Evidence
- The clustering method resulted in an average RMSD quantization error of 1.19 Å for 27 structural representatives.
- The area under the ROC curve for detection of the 27 classes ranged from 0.68 to 0.88.
- Using profiles based on amino-acid properties improved prediction accuracy.
Takeaway
The researchers created a new way to predict how parts of proteins will fold based on their sequences, which can help scientists understand proteins better.
Methodology
The method involves clustering local structures and training a discriminative model to predict local structure representatives from local sequence information.
Potential Biases
The probability estimates may be biased towards the dominant class due to class imbalance.
Limitations
The method may struggle with longer protein fragments due to increased structural variation.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website