Comparing T cell Epitope Prediction Methods for HIV Proteins

The application of computer-driven algorithms for T cell epitope prediction in the context of HIV protein gp160 illustrates the advantages over the traditional overlapping peptide method. To map T cell epitopes for a single clade or laboratory strain’s gp160 protein, approximately 60 overlapping 20-mers need to be generated, as depicted in Figure 1. However, employing algorithms like EpiMer and EpiMatrix streamlines the process by focusing on the 10 to 15 peptides most likely to bind to MHC molecules and present at the antigen-presenting cell surface. While this strategy may overlook some epitopes, it expedites the resolution of critical research queries associated with HIV vaccine development.

Table 1. Comparison of the sensitivity of the Overlapping Peptide method, AMPHI, and EpiMer ML 0595 for the prediction of HIV epitopes

Comparison of Methods Overlapping AMPHI EpiMer
Weighted Mean Percent Efficiency* 48% 52% 63%
range 43% – 100% 37% – 68% 61% – 64%
Weighted Mean Percent Sensitivity** 100% 69% 65%
range 100% – 100% 33% – 93% 22% – 86%
Mean Sensitivity per amino acid 2.7 3.8 4.9
range 0.6 – 6.4 1.0 – 8.3 1.3 – 8.1
Mean Æ Sensitivity per AA 1.6 2.4

Source: Roberts et al. AIDS Research and Human Retroviruses, May 1995, for details9. The summary highlights the predictive power of EpiMer, AMPHI, and the overlapping peptide simulation for the four HIV-1 proteins.

The table demonstrates the superior performance of the EpiMer ML 0595 algorithm over the overlapping method and the AMPHI method for predicting T cell epitopes. Both the overlapping method and the AMPHI method require more peptides and amino acids, resulting in EpiMer’s 2.4-fold higher sensitivity in detecting T cell epitopes per synthesized amino acid. Notably, the sensitivity of the method decreases with the size of the protein. For smaller proteins like tat, the sensitivity and specificity per amino acid were comparable across the three methods.

Several additional MHC-binding motif-based algorithms, such as those by Parker et al., Altuvia et al., DeLisi et al., and Brusic et al., have been described. However, with the exception of EpiMer, none of these algorithms have undergone in vitro testing, nor do they offer the clustering option. Preliminary studies of EpiMer peptides strongly support the utility of MHC-motif-based predictions for T cell epitope identification and the clustering concept. Notably, there exists a close correlation between the number of individuals in an outbred human population responding to the clustered peptides and the number of predicted motifs contained within the peptides.