Reliable B Cell Epitope Predictions: Impacts of Method Development and Improved Benchmarking

We present an updated version of the B-cell epitope prediction method; DiscoTope, that on the basis of a protein structure and epitope propensity scores predicts residues likely to be involved in B-cell epitopes

Jens Vindahl Kringelum; Claus Lundegaard; Ole Lund; Morten Nielsen


Scholarcy highlights

  • The interaction between antibodies and antigens has been the center of attention for multiple disciplines within immunological research and applications, and a dozen of methods for computational mapping of antibody binding on the antigen surface have been developed in the later years
  • One of the most important immune system events involved in clearing infectious organisms is the interaction between the antibodies and antigens
  • We demonstrate that the low performances to some extent can be explained by poorly defined benchmarks, and that inclusion of additional biological information greatly enhances the predictive performance
  • This suggests that, given proper benchmark definitions, stateof-the-art B cell epitope prediction methods perform significantly better than generally assumed
  • As neither the definition of spatial neighborhood nor surface measures are trivial tasks, one aim of the presented work was to investigate the ability of a new scoring function for defining a spatial neighborhood and different surface measures to improve the accuracy for B-cell epitope prediction
  • Using the benchmark data set from the original DiscoTope paper, we demonstrate that the updated method has a significantly increased predictive performance
  • In difference to the function proposed by Sweredoski and Baldi, which uses 5 distance thresholds to stepwise decrease the weight on log-odds ratios, the function proposed here is defined by only two parameters: a sequential smoothing window w and a distance scale kps

Need more features? Save interactive summary cards to your Scholarcy Library.