Antibody-protein interactions: benchmark datasets and prediction tools evaluation
2007

Evaluating Antibody-Protein Interaction Prediction Tools

Sample size: 144 publication Evidence: low

Author Information

Author(s): Julia V. Ponomarenko, Philip E. Bourne

Primary Institution: San Diego Supercomputer Center, University of California, San Diego

Hypothesis

Can we establish a benchmark for predicting B-cell epitopes from antibody-protein complexes using 3D structures?

Conclusion

The study found that existing methods for predicting B-cell epitopes performed poorly, with no method exceeding 40% precision.

Supporting Evidence

  • Two benchmark datasets were created from 3D structures of antibody-protein complexes.
  • Performance of eight prediction methods was evaluated, with none exceeding 40% precision.
  • The area under the ROC curve for evaluated methods was about 0.6 for some methods.

Takeaway

This study looked at how well different computer programs can guess where antibodies stick to proteins, and found that they don't do a very good job.

Methodology

The study evaluated eight web-servers for predicting B-cell epitopes using benchmark datasets derived from 3D structures of antibody-protein complexes.

Potential Biases

The methods evaluated may not generalize well due to the limited number of available antibody-protein complex structures.

Limitations

The performance of the methods was generally poor, and the study suggests that the generality of antigenicity may hinder accurate predictions.

Statistical Information

P-Value

p<0.05

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1186/1472-6807-7-64

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication