Fig. 1. B*5101 binding threshold. Comparison between the score predicted by the BIMAS algorithm for random peptides and experimentally measured peptides (six databases). The cutoff presented in the drawing is for the combined maximal inclusion of the presented peptides and exclusion of the random peptides. As can be seen, even for this extreme threshold (taking into account the large number of false positives in existing databases), there is a clear distinction between the random peptides upper cutoff and experimentally observed peptide lower cutoff. This is true for the vast majority of alleles. The lower part of the drawing represents the fraction of false positives and false negatives for each cutoff. The optimal cutoff was designed as the one minimizing both false positives and false negatives.

Fig. 2. T cell epitopes prediction for the HCV virus. Starting from upper-right corner A: the HCV polyprotein is cleaved to 10 proteins (composing 11,754 possible peptides (nonamers)). B: From those 2173 are computed to be cleaved by the proteasome. C: 1164 peptides are then computed to pass through the TAP machinery into the ER. In the ER, D: 6324 (1164 single repetition) peptides binds to different HLA-I molecules (29 alleles). E: Only 2896 peptides (523) from the binding peptides are 90% conserved. F: None of them is similar to self peptides. These peptides are candidate for a peptide-vaccine. The algorithms can thus be summarized by the following six steps. A: Analysis of the full genome and the decomposition of the genome into all possible nonamers. B: Computation of the probability that each peptide should be cleaved. C: Computation of the TAP binding probabilities. D: For each separate HLA calculation of the MHC-peptide binding probability. E: Analysis of the conserved peptides. F: Comparison to self peptide. At each stage, a cutoff was used and only peptide with a supra-cutoff score were chosen.

Fig. 3. Different proteins present different number of epitopes. Each column represents the SIR score deviation from the average SIR score ((SIR − average SIR)/average SIR). The average SIR score is 0.174 epitopes per protein. Only columns E2 and NS4B had a significant result at the 0.05 level. A similar result was obtained for the mouse analysis.

Fig. 4. The optimal vaccine sequence length. The genetic algorithm was used with different lengths (different number of peptides). The X-axis represents the number of peptides used (5–60 peptides) and the Y-axis represents the optimal score obtained for each combination length. The algorithm maximizes a score (blue diamonds) which is a combination of four components: (A) the fraction of cleaved peptides, which always saturate on 1 and are thus not shown, (B) the number of proteins covered (black squares), (C) the number of HLA alleles covered (red triangles) and (D) the number of HLA/protein combinations covered (green x-points). Beyond a length of 25 peptides, the score saturates, while the score length keeps decreasing. We thus limited the number of peptides to 25. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

Fig. 5. Genetic algorithm convergence. The X-axis represents the number of generation (log scale). The algorithm maximizes four components which are combined in the total score (blue diamonds): (A) the number of cleaved peptides (light blue plus-sign), (B) the number of proteins covered (black squares), (C) the number of HLA alleles covered (red triangles) and (D) the number of HLA/protein combinations covered (green x-points). Each of the parameters is normalized to the maximal possible score. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

Fig. 6. The HCV sequence vaccine. A combination of 25 peptides which yield an optimal sub-unit peptide vaccine. The marked amino acids represent the flanking regions of each peptide.

Fig. 7. The sequence vaccines against (A) HCV virus for mice cells, (B) HIV-I M clad B, (C) Flu H3N2 and (D) H5N1virus, respectively. The sequences for the HIV and Flu vaccines are based on a specific subtype. As a result, the optimal sequences of the HIV and the Flu have covered all the possible alleles (29 alleles) and proteins (8 segments for the Flu and 9 proteins for the HIV). The underlined epitopes in HIV overlap with known epitopes.

MHC-I table

This table contains all the required information on each of the HLA alleles used in the current analysis. For each HLA allele the table contains its locus (column A), its name (column B), the name used for it in the BIMAS tables (column C). When we do not mention a name the full name was used. Its average frequency in the human population (column D), the constant provided by BIMAS to calibrate their tables (column E) and the optimal threshold computed (column F).