Table 1

Predictive performance for the 14 HLA-DR alleles in the quantitative IEDB benchmark dataset.

Allele

#

TEPITOPE

SMM-align

NN

NN-W

NN-P1

NN-W-P1

NN-xPFR


DRB1*0101

5166

0.720

0.802

    0.837

0.836

0.833

0.833

0.813

DRB1*0301

1020

0.664

0.795

0.808

0.816

    0.817

    0.817

0.778

DRB1*0401

1024

0.716

0.751

    0.768

0.771

0.764

0.766

0.764

DRB1*0404

663

0.770

0.801

0.815

    0.818

0.816

0.817

0.814

DRB1*0405

630

0.759

    0.789

0.771

0.781

0.779

0.784

0.773

DRB1*0701

853

0.761

0.812

    0.844

0.841

0.839

0.838

0.813

DRB1*0802

420

0.766

0.787

0.826

    0.832

0.820

0.822

0.802

DRB1*0901

530

    0.655

0.623

0.616

0.618

0.640

0.623

DRB1*1101

950

0.721

0.796

0.822

    0.823

0.818

0.813

0.802

DRB1*1302

498

0.652

0.785

0.822

    0.831

0.822

0.824

0.811

DRB1*1501

934

0.686

0.727

0.754

0.758

0.754

0.754

    0.761

DRB3*0101

549

0.836

    0.855

0.844

0.838

0.841

0.831

DRB4*0101

446

0.793

0.811

0.811

0.815

    0.818

0.813

DRB5*0101

924

0.680

0.761

0.789

    0.797

0.790

0.790

0.787


Ave

0.718

0.778

0.796

0.798

0.795

0.797

0.785

Ave*

0.716

0.785

0.809

0.810

0.807

0.808

0.793


The performance was estimated using five-fold cross-validation and is given as the area under the ROC curve (AUC) calculated using a binding affinity threshold of 500 nM. SMM-align is the SMM-align method described by Nielsen et al. [20] re-trained on the quantitative benchmark data set. TEPITOPE refers to the method developed by Sturniolo et al. [17]. NN is the standard NN-based method, NN-W is the NN-based method including data redundancy step-size rescaling, NN-P1 is the NN-based method including PSSM-P1 amino acid encoding, NN-W-P1 is the NN-based method including data redundancy step-size rescaling and PSSM-P1 amino acid encoding. NN-xPFR is the NN-W-P1 method excluding peptide-flanking residue encoding. For each allele, the best performing NN method is highlighted in bold and the best performing of all methods is underlined. Ave is the average per-allele performance over all 14 alleles, and Ave* is the average predictive performance over the 14 alleles weighted by the number of data points for each allele.

Nielsen and Lund BMC Bioinformatics 2009 10:296   doi:10.1186/1471-2105-10-296

Open Data