Consistent probabilistic outputs for protein function prediction
2008

Improving Protein Function Prediction with Reconciliation Methods

Sample size: 2000 publication Evidence: high

Author Information

Author(s): Guillaume Obozinski, Gert Lanckriet, Charles Grant, Michael I. Jordan, William Stafford Noble

Hypothesis

Can reconciliation methods improve the consistency and accuracy of protein function predictions derived from independent classifiers?

Conclusion

Isotonic regression is the most effective reconciliation method for protein function prediction, consistently yielding high precision across various evaluation modes.

Supporting Evidence

  • Isotonic regression generally outperformed other reconciliation methods across various evaluation modes.
  • Many reconciliation methods resulted in lower precision than the original predictions.
  • The study identified that isotonic regression can leverage the structure of the Gene Ontology to improve classification.

Takeaway

This study shows that when predicting what proteins do, using a method that makes sure predictions make sense together can help scientists get better answers.

Methodology

The study employed support vector machines (SVMs) for initial predictions and applied various reconciliation methods to improve consistency with Gene Ontology.

Potential Biases

Some reconciliation methods may yield lower precision than unreconciled estimates, potentially misleading interpretations.

Limitations

The performance of reconciliation methods can vary significantly based on the specific terms and contexts, particularly for smaller terms.

Statistical Information

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1186/gb-2008-9-s1-s6

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication