05.08.2013 Views

New Approaches to in silico Design of Epitope-Based Vaccines

New Approaches to in silico Design of Epitope-Based Vaccines

New Approaches to in silico Design of Epitope-Based Vaccines

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

5.4. EXPERIMENTAL RESULTS 59<br />

Let G be the set <strong>of</strong> genotypes with<strong>in</strong> the population <strong>of</strong> <strong>in</strong>terest and p(g) the probability<br />

<strong>of</strong> genotype g. Furthermore, let η(E, g) be the number <strong>of</strong> epi<strong>to</strong>pes <strong>in</strong> an epi<strong>to</strong>pe set E<br />

which are immunogenic with respect <strong>to</strong> an MHC allele <strong>in</strong> g. The average number <strong>of</strong> active<br />

epi<strong>to</strong>pes per <strong>in</strong>dividual <strong>in</strong> the population results from<br />

ψ(E) = <br />

p(g) η(E, g). (5.5)<br />

g∈G<br />

These non-l<strong>in</strong>ear requirements cannot be <strong>in</strong>corporated <strong>in</strong><strong>to</strong> the ILP directly. It is,<br />

however, possible <strong>to</strong> search a sufficiently large set <strong>of</strong> optimal and suboptimal solutions for<br />

the best set <strong>of</strong> epi<strong>to</strong>pes that displays the required properties. Furthermore, for some nonl<strong>in</strong>ear<br />

requirements, l<strong>in</strong>ear substitute requirements may be found: e.g., requir<strong>in</strong>g a certa<strong>in</strong><br />

degree <strong>of</strong> population coverage is equivalent <strong>to</strong> requir<strong>in</strong>g a certa<strong>in</strong> degree <strong>of</strong> allele coverage.<br />

It has <strong>to</strong> be noted that, while ILPs require constra<strong>in</strong>ts <strong>to</strong> be l<strong>in</strong>ear, the non-l<strong>in</strong>earity <strong>of</strong> a<br />

constra<strong>in</strong>t does not necessarily imply that the correspond<strong>in</strong>g optimization problem cannot<br />

be solved efficiently. The solver CPLEX [53], for example, can handle <strong>in</strong>teger programs<br />

with quadratic constra<strong>in</strong>ts. Furthermore, heuristics can be employed <strong>to</strong> determ<strong>in</strong>e nearoptimal<br />

solutions <strong>to</strong> such problems. For more <strong>in</strong>formation on optimally solv<strong>in</strong>g <strong>in</strong>teger<br />

non-l<strong>in</strong>ear programs please refer <strong>to</strong> [107, 108].<br />

5.4 Experimental Results<br />

In order <strong>to</strong> show the effectiveness <strong>of</strong> the proposed approach, we compare our optimal<br />

strategy (best overall immunogenicity, BOI) with two simple approaches:<br />

• randomly select k peptides out <strong>of</strong> a pool <strong>of</strong> good epi<strong>to</strong>pes (random set <strong>of</strong> epi<strong>to</strong>pes,<br />

RSE) and<br />

• a simple greedy approach: pick the k best epi<strong>to</strong>pes from the set (best epi<strong>to</strong>pe-wise<br />

immunogenicity, BEI).<br />

As performance measures the theoretical ga<strong>in</strong> <strong>in</strong> overall immunogenicity as well as the<br />

number <strong>of</strong> epi<strong>to</strong>pes required <strong>to</strong> achieve a similar overall immunogenicity are employed.<br />

The three epi<strong>to</strong>pe selection strategies were used <strong>to</strong> select different-sized sets <strong>of</strong> peptides<br />

from a set <strong>of</strong> 4,461 HCV n<strong>in</strong>emers. (The data set is described <strong>in</strong> more detail <strong>in</strong> Section 7.2.)<br />

For BOI, the basic ILP 5.1 was used <strong>to</strong> maximize overall immunogenicity. BEI selects the<br />

epi<strong>to</strong>pes with the highest sum <strong>of</strong> immunogenicities irrespective <strong>of</strong> the probabilities <strong>of</strong> the<br />

correspond<strong>in</strong>g MHC alleles.<br />

The overall immunogenicity <strong>of</strong> each epi<strong>to</strong>pe set was determ<strong>in</strong>ed and is displayed <strong>in</strong><br />

Figure 5.2. For RSE, mean and standard deviation <strong>of</strong> 100 random selections <strong>of</strong> differentsized<br />

epi<strong>to</strong>pe sets from the 100 most immunogenic peptides are shown. The BEI curve<br />

shows sudden <strong>in</strong>creases <strong>in</strong> overall immunogenicity from 0 <strong>to</strong> 1, 10 <strong>to</strong> 13, and from 20<br />

<strong>to</strong> 21 epi<strong>to</strong>pes. This is caused by the selection <strong>of</strong> epi<strong>to</strong>pes that are highly immunogenic<br />

with respect <strong>to</strong> HLA-A*02:01, which is the most common among the considered alleles<br />

(pℓ = 0.145). All other selected epi<strong>to</strong>pes are highly immunogenic with respect <strong>to</strong> less

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!