Performance comparison between SVRMHC models and "additive method" models

    We built “additive method” models for the 42 MHC molecules as described in (Doytchinova, et al., 2002; Doytchinova and Flower, 2003) , with the same dataset used to construct corresponding SVRMHC models. The 36 class I SVRMHC models produced an average cross-validated q2 of 0.414, compared to 0.294 for the 36 corresponding “additive” models. Thus, cross-validated q2 values for the SVRMHC models indicate a significant improvement over corresponding “additive” models (P<0.0001, Wilcoxon rank sum test). Next, we determined and removed outliers as described in (Doytchinova and Flower, 2002; Liu, et al., 2006) . After outlier removal, all models improved. The average cross-validated q2 of the 36 class I SVRMHC models was 0.471, in comparison to 0.487 for the 36 class I “additive” models. After outlier removal, cross-validated q2 values for the SVRMHC and “additive” models do not differ significantly (P=0.32, Wilcoxon rank sum test). The average number of outliers determined and removed by the SVRMHC models was 2.0, compared to 7.1 for the “additive method” models. This suggests that without removing outliers, class I SVRMHC models out perform class I “additive” models. After removal of outliers, SVRMHC models and “additive” models offered comparable performance, though a smaller number of outliers were removed for the SVRMHC models.

    The 6 class II SVRMHC models produced an average cross-validated r of 0.60, in comparison to 0.39 produced by the 6 class II “additive” models. The cross-validated r values of the class II SVRMHC models were significantly higher than those for the class II “additive” models (P=0.028, Wilcoxon rank sum test).

Table 1 Performance comparison of class I additive method models and SVRMHC models, with and without removal of outliers. All models are nomamer models unless marked otherwise.

Table 2 Performance comparison of class II additive method models and SVRMHC models. All models are nomamer models.

References

Doytchinova, I.A., Blythe, M.J. and Flower, D.R. (2002) Additive method for the prediction of protein-peptide binding affinity. Application to the MHC class I molecule HLA-A*0201, J Proteome Res , 1 , 263-272.

Doytchinova, I.A. and Flower, D.R. (2002) Physicochemical explanation of peptide binding to HLA-A*0201 major histocompatibility complex: a three-dimensional quantitative structure-activity relationship study, Proteins , 48 , 505-518.

Doytchinova, I.A. and Flower, D.R. (2003) Towards the in silico identification of class II restricted T-cell epitopes: a partial least squares iterative self-consistent algorithm for affinity prediction, Bioinformatics , 19 , 2263-2270.

Liu, W., Meng, X., Xu, Q., Flower, D.R. and Li, T. (2006) Quantitative prediction of mouse class I MHC peptide binding affinity using support vector machine regression (SVR) models, BMC Bioinformatics , 7 , 182.

Send questions and comments to SVRMHC@biolead.org.
Copyright 2006-13 Biolead.org. All rights reserved.