METHODOLOGY

The algorithm used to derive the prediction model is described in Bioinformatics, Vol. 17 no. 10, 2001, “Predicting class II MHC/peptide multi-level binding with an iterative stepwise discriminant analysis meta-algorithm” by R.R. Mallios. The motif matrix was derived from datasets retrieved from the MHCBN website at http://www.imtech.res.in/raghava/mhcbn/ . [Bhasin, M., Singh, H., Raghava, G. P. S.(2002) MHCBN:A Comprehensive Database of MHC Binding and Non-Binding Peptides.Nucleic Acids Research, (online) http://www3.oup.co.uk/nar/database/summary/180 .]

Let binding levels be denoted by 0 = none, 1 = low, 2 = moderate and 3 = high. To calculate the predicted probability that a given nonamer binds an HLA-DRB1 molecule with affinity level i, first calculate u₀, u₁, u₂ and u₃, where the u_i = constant_i + b_1i + b_2i + b_3i+ b_4i + b_5i+ b_6i + b_7i + b_8i + b_9i, and b_ji is the element of the motif matrix which corresponds to the residue in the jth position. Then calculate P_i = e^uⁱ /(e^u⁰+ e^u¹ + e^u² + e^u³).

Residue		Position
		1	2	3	4	5	6	7	8	9
Constant	Level
-4.55	0
-17.16	1
-11.95	2
-15.46	3
A	0		1.28	0.76	1.94		1.97		2.22
	1		1.89	3.07	4.49		1.94		2.67
	2		4.00	-0.22	3.12		3.12		4.54
	3		3.91	1.24	2.61		4.63		3.48
C	0	2.32						1.88	2.00
	1	2.24						12.08	2.18
	2	8.89						0.58	3.47
	3	3.03						-0.76	-0.62
D	0						2.06
	1						0.74
	2						0.55
	3						-0.71
E	0		0.95		2.06		2.71			1.52
	1		2.87		5.24		5.91			-0.38
	2		-0.41		2.59		2.77			2.67
	3		-0.22		2.35		1.88			0.84
F	0	2.38						1.49
	1	1.93						4.99
	2	3.59						2.32
	3	17.88						2.42
G	0	2.37