Information Technology Reference
In-Depth Information
6 Analysis of Coverage
We have seen in the last section that self clustering drives increased coverage and an
increase in the number of detectors made. Antigen clustering also increases coverage
but decreases the number of receptors made. In this section we take a closer look at
these results, to examine the strategies that our AIS employs for covering antigens
under different cluster arrangements. For example, is antibody clustered, and how
does this clustering change according to environment? In order to answer this
question, we analysed representative antibody populations taken from different points
in the self cluster/antigen cluster space. In each case we analyzed one representative
individual and compared this against (averaged) random performance.
Table 3 shows the number of antibodies produced for each point in cluster space.
Interestingly, these figures contain no duplicate detectors. Greater numbers of
antibodies are produced by the individuals that use 2 or 3 gene libraries. The number
of antibodies created increases with the number of self clusters and decreases with the
number of antigen clusters.
Table 3.
Number of antibodies produced by gene library individuals for different points in
cluster space (data shown graphically in figure 5). The biggest number of antibodies produced
for each point in cluster space is shown in bold; for each gene library configuration by
underlining. For comparison, random creation (table 2) consistently produces <25 antibodies.
Description
1lib
2 libs
3 libs
0self - 0antigens
43
261
170
0self - 5antigens
38
157
94
2self- 2 antigens
124
361
441
5self - 0antigens
120
490
669
5self - 5antigens
88
343
579
We were interested in seeing how coverage compared against the theoretical
optimum and a random creation strategy. The latter is easy to test - we just randomly
create antibodies (discarding duplicates) until we get the same number that the gene
libraries produce. The former is more difficult, but fortunately Wierzchon [12] has
shown how this is possible. We used his paper to code an algorithm, the pseudocode
of which is shown in Figure 6.
Figure 7 shows that, in general, coverage increases as the antigen clustering
increases. Use of one library consistently outperforms random antibodies; two and
three libraries require a clustered space to do so. It is important to bear in mind that
random
here refers to the same number of antibodies; as 2 and 3 libraries produce
large numbers of antibodies (see table 3), then the same number of (randomly
produced) antibodies will of course give high coverage. The cost of creation is not
taken into account here, as it is dealt with in Section 3. As reported in section 5, the
best coverage is achieved with highly clustered self and antigens.
