# Data sets (.txt files) for each allele are provided as peptide sequence followed by the IC50 of the
# peptide-MHC complex (in nM) and the concentration of reporter peptide (in nM if reported or "NA" if
# not reported), all in tab-delimited columns. Two data sets are provided per allele: as downloaded
# from AntiJen and after homologous sequences are filtered out by UniqueProt.

# The sizes of the data sets (before/after filtering by UniqueProt) are as follows:
#	DRB1*0101 (464/303)
#	DRB1*0401 (606/414)
#	DRB1*0404 ( 81/ 54)
#	DRB1*0405 (116/102)
#	DRB1*1501 (343/213)

# Two postscript files for each allele are also provided: the length distribution of peptides binding
# each allele and the local regression fit made by R.

# ProPred predictions from which the AROC scores in Table 3 were calculated are also included as 
# text files. The left-hand column shows experimentally derived pIC50 values, while the right-hand
# column shows predictions made by each algorithm. For each allele and algorithm, a separate file is
# provided. The file "DRB1_0101_ProPred_combiRule.txt" gives the predictions made on the sequences
# in AntiJen for DRB1*0101 (after filtering by UniqueProt) when the combination rule of Doytchinova and
# Flower (2003) was used on the ProPred 9mer predictions.

# Finally alternative versions of Tables 2 and 3 in the manuscript are provided. These versions use the
# Pearson correlation coefficient to score algorithm performance (as opposed to the area under receiver
# operating characteristic curve, or AROC, which is used in the manuscript).