Flanking p10 contribution and sequence bias in matrix based epitope prediction: revisiting the assumption of independent binding pockets

Table 1 Summary of analysis

A. Full data
Data set	method	-m	J	r ²
GAD65	P9	0.352	30.22	0.373
	P10	0.341	26.75	0.445
	PP10	0.383	22.37	0.536
ICA69	P9	0.338	29.91	0.325
	P10	0.275	33.24	0.250
	PP10	0.327	29.27	0.340
VZV	P9	0.252	8.16	0.386
	P10	0.225	8.99	0.324
B. Data edited to remove poor binders
Data set	method	-m	J	r ²
GAD65	P9	0.324	25.44	0.279
	P10	0.368	20.12	0.429
	PP10	0.410	16.12	0.543
ICA69	P9	0.308	24.27	0.251
	P10	0.244	25.47	0.214
	PP10	0.308	21.22	0.345
VZV	P9	0.334	1.92	0.573
	P10	0.324	2.56	0.431

The fit of prediction to binding has been calculated for each prediction method using measures of slope (-m), sum of the square terms of the residual error (J) and r-square value (r²) for peptide sequences from GAD65, ICA69 and VZV. Table 1A shows the statistics for the unedited data, and Table 1B shows the equivalent statistics for the data edited to remove poor binding measurements.

ISSN: 1472-6807