Improving T-Cell Epitope Prediction by Avoiding Rescaling
Author Information
Author(s): Aidan MacNamara, Ulrich Kadolsky, Charles R. M. Bangham, Becca Asquith
Primary Institution: Department of Immunology, Imperial College School of Medicine, London, United Kingdom
Hypothesis
Does rescaling remove genuine biological variation from predicted affinities when comparing predictions across MHC molecules?
Conclusion
Removing the condition of rescaling improves the prediction software's performance and leads to more accurate epitope predictions.
Supporting Evidence
- Rescaling led to a significant loss of performance in epitope prediction accuracy.
- Non-rescaled predictions identified more true epitopes with fewer false positives.
- The study demonstrated that rescaling obscures crucial biological variations in MHC binding affinities.
Takeaway
The study found that adjusting prediction scores to make them comparable can actually hide important differences between MHC molecules, making predictions less accurate.
Methodology
The study tested two prediction software packages, NetCTL and NetMHC, to compare the effects of rescaling on epitope prediction accuracy.
Limitations
The study primarily focused on two prediction methods and may not generalize to all epitope prediction algorithms.
Statistical Information
P-Value
p<0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website