Benchmarking Predictions of Peptide Binding to MHC-I Molecules
Author Information
Author(s): Peters Bjoern, Bui Huynh-Hoa, Frankild Sune, Nielsen Morten, Lundegaard Claus, Kostem Emrah, Basch Derek, Lamberth Kasper, Harndahl Mikkel, Fleri Ward, Wilson Stephen S, Sidney John, Lund Ole, Buus Soren, Sette Alessandro
Primary Institution: La Jolla Institute for Allergy and Immunology
Hypothesis
Can different prediction methods accurately identify peptide binding to MHC class I molecules?
Conclusion
The study found that an artificial neural network outperforms matrix-based prediction methods for peptide binding to MHC class I molecules.
Supporting Evidence
- The dataset includes 48,828 quantitative peptide-binding affinity measurements.
- The neural network method showed better generalization on small datasets compared to matrix-based methods.
- Tools based on combinatorial peptide libraries performed well in binding predictions.
- The study provides a benchmark for future prediction method evaluations.
- ANN predictions were significantly better than ARB and SMM methods.
- Data from two different assay systems were used to create a more homogeneous dataset.
- Predictions were validated using five-fold cross-validation.
Takeaway
This study created a big list of how well different peptides stick to MHC molecules, helping scientists figure out which prediction methods work best.
Methodology
The study involved collecting peptide-binding affinity measurements and comparing the performance of three prediction methods using five-fold cross-validation.
Potential Biases
The reliance on data from only two sources may introduce bias in the predictions.
Limitations
The dataset is limited to two assay systems, which may not represent all possible peptide-MHC interactions.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website