FEATURE Framework for Protein Function Annotation
Author Information
Author(s): Inbal Halperin, Dariya S Glazer, Shirley Wu, Russ B Altman
Primary Institution: Stanford University
Hypothesis
Can the FEATURE framework improve the annotation of protein functions using structural and functional modeling?
Conclusion
The FEATURE framework effectively models molecular functions without relying on significant sequence or fold similarity, enhancing functional coverage and efficiency.
Supporting Evidence
- FEATURE can automatically generate training sets from various sources to improve functional coverage.
- The framework has been successfully applied to predict functional sites in proteins with low sequence similarity.
- FEATURE's performance improves when coupled with molecular dynamics simulations to account for protein dynamics.
Takeaway
FEATURE is a tool that helps scientists understand what proteins do by looking at their shapes and structures, even when they don't look like any proteins we already know.
Methodology
The FEATURE framework uses a combination of structural and functional modeling, including supervised machine learning and physicochemical property calculations, to predict protein functions.
Potential Biases
The selection of training sets, particularly negative sites, can introduce bias into the model's predictions.
Limitations
The framework may struggle with proteins that have very low sequence identity to known proteins, and the choice of negative training sites can influence model performance.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website