Related Experiment Videos
Evaluation of structural and evolutionary contributions to deleterious mutation prediction
Christopher T Saunders1, David Baker
1Department of Genome Sciences, University of Washington, Seattle 98195, USA.
Journal of Molecular Biology
|September 25, 2002
Summary
Predicting harmful protein mutations is improved by combining structural and evolutionary data. Integrating structural features like C(beta) density with evolutionary scores like SIFT enhances prediction accuracy, especially with limited homologous sequences.
Area of Science:
- Computational Biology
- Protein Structure Prediction
- Bioinformatics
Background:
- Automated prediction of deleterious protein mutations often uses structural and evolutionary information.
- The relative importance of structural versus evolutionary factors in mutation prediction is not well understood.
- Accurate prediction of mutation effects is crucial for understanding genetic diseases and protein function.
Purpose of the Study:
- To evaluate the individual and combined contributions of structural and evolutionary features for predicting deleterious protein mutations.
- To develop and test simple deleterious mutation models using various structural and evolutionary metrics.
- To determine the optimal combination of features for accurate mutation effect prediction across different data types.
Main Methods:
- Developed predictive models using structural features (e.g., solvent accessibility, C(beta) density) and evolutionary scores (SIFT).
- Tested models on experimental mutagenesis data and human allele datasets.
- Employed classification trees and cross-validation to assess prediction accuracy and error rates.
Main Results:
- The combination of C(beta) density and SIFT score yielded the highest prediction accuracy (20.5% error) on experimental data.
- Structural information, particularly C(beta) density, significantly improved predictions when fewer homologous sequences were available.
- Even noisy C(beta) terms from ab initio predicted structures showed improvement when combined with SIFT for limited homologs.
Conclusions:
- Integrating structural information is recommended for deleterious mutation prediction when fewer than five to ten homologous sequences are available.
- Ab initio predicted protein structures may offer a viable alternative when high-resolution structures are absent.
- Combined structural and evolutionary approaches provide superior accuracy for predicting the functional impact of protein mutations.