Information Technology Reference
In-Depth Information
Wierzchon's algorithm also allows us to measure degree of targeted coverage. So,
for example, the number of antigen templates covered (Figure 9a) show how
completely the antigens are matched, while the number of antibody templates
covering (Figure 9b) indicate how 'targeted' the antibody set is at that particular
antigen population. Again, templates covered (9a) for one library is close to or above
'random' behaviour, while two and three libraries cover far fewer antigen templates,
even when delivering superior coverage (for example, 2 libraries with 0self5antigens).
The targeting of antibodies (fig 9b) shows a higher than random focusing of templates
only in tightly clustered scenarios (eg 2lib2antigens); this is consistent with an
antibody population tailored to the fixed antigen set.
7 Conclusions
Gene libraries clearly introduce initialisation bias to antibody creation. We have
shown that such bias induces superior coverage, but that this improvement is not
purely through reducing the cost of negative selection, nor of combinatorial effects;
rather some antigen mapping must be occurring. For unclustered antigens, the
antibodies generated retain high diversity; as the antigens become more clustered the
antibody population becomes less diverse (fig 8) and more targeted (fig 9). The
comparable (even superior) performance of 2 libraries (versus 1 library) is also
compelling given the combinatorial advantages. In summary, we suggest that gene
libraries: provide combinatorial efficiency; improve coverage; reduce the cost of
negative selection; and allow targeting of fixed antigen populations.
We have chosen to analyse gene libraries by assessing their effect on established
AIS notions of negative selection and coverage. However, gene libraries will also
have an impact on other immune metaphors such as homeostasis [13] and danger
[14], and these topics would be interesting directions for future work. Representation
other than bit-strings, and mapping operators other than rContiguous bits, would also
be suitable subjects for further work. Dealing with co-evolving antigens is another
topic for further study.
For now, we conclude that gene libraries do appear to produce a tangible benefit in
a defined space, we suggest a mechanism whereby they achieve this, and present a
method for analysing their performance.
References
1.
Steve Cayzer, Jim Smith, James Marshall & Tim Kovacs (2005) What have gene libraries
done for AIS? Proceedings ICARIS-2005, 4th International Conference on Artificial
Immune Systems, LNCS 3627, pp 86-99, Springer-Verlag, Banff, Canada.
2.
Goldsby, R.A., Kindt, T.J., Osborne, B.A., Kuby, J.: “Immunology” W H Freeman, New
York 5th edition (2003)
3.
Matzinger, P. The Danger Model: A renewed sense of self. Science 296 (2002) 301-305
4.
Perelson, A.S., Hightower, R., Forrest, S. (1996) “Evolution and somatic learning in
V-region genes. Research in Immunology 147 pp.202-208
