Andrew, A.M. (2004), "Immunocomputing: Principles and Applications", Kybernetes, Vol. 33 No. 8, pp. 1332-1333. https://doi.org/10.1108/03684920410545315
Emerald Group Publishing Limited
Copyright © 2004, Emerald Group Publishing Limited
This is an exploration of the possibility of basing new computing strategies on the immune system of the body, in something the same way as artificial neural nets have become an accepted technique.
It is argued that the immune system performs complex information‐processing tasks, including pattern recognition and self‐organisation, and is better understood than the nervous system. As these authors say in their introduction:
Actually, the immune system possesses not only memory, but is also able to learn, to recognise, and to make decisions about how to treat any molecule (antigen) even if that molecule has never before existed on Earth. Besides, the immune system defines the great diversity of possible molecular shapes in the highly individual context of its own experience, and it has developed strategies to free the genome from the task of straight‐coding such diversity.
The requirement for pattern recognition requires the means of finding matching substrings within the strings constituting protein molecules or their analogues. This has been achieved by conventional computing methods in the search facilities associated with, for example, the human genome project. However, it is hoped that in an immunocomputer the result would be achieved in what seems to be the natural method. The data string represented by a protein molecule determines folding of the physical string in complex ways so as to arrive at an overall shape. Interactions with other molecules are, at least in part, determined by this shape.
It is suggested that here there is the basis of a negative‐selection algorithm that would allow what is termed self‐nonself discrimination of the kind that must underlie the body's immunological response. Reference is made to a project in New Mexico to protect a computer system by such means against unauthorised use, corruption of data files, and viruses.
In the second chapter, the mechanics of protein folding, termed self‐assembly, are analysed in considerable detail, depending on rather complex three‐dimensional geometry. The spatial conformation of the protein is determined by the solid angle between successive bonds, and this in turn is partly determined by the types of atom bonded but also has some freedom. The latter allows the molecule to settle to different forms, corresponding to energy minima, so that there is memory. The energy is affected by neighbouring molecules and there is recognition and either binding or repulsion.
The building blocks of an immunocomputer are formal proteins, or FPs, having similar status to formal neurons. FPs may be free or may be receptors of formal cells corresponding to lymphocytes and of two varieties, B‐cells and T‐cells. These are able to change their bindings and can either proliferate or die according to the bindings they form.
It is demonstrated that a computer formed of FPs and formal cells would allow pattern recognition and both supervised and unsupervised learning, and could be extended to language representation and knowledge‐based reasoning. It would also have advantages in the modelling of natural and technical systems, and here reference is made to a paper by Tarankov and Adamatzky (2002) where the three‐dimensional geometry that was invoked in connection with protein folding is applied to the folding, or draping, of cloth. The published paper includes only a passing reference to immunocomputing as such but is an interesting application of the general approach.
Further possible applications that are discussed include, perhaps not surprisingly, some of an epidemiological nature, and others in the nature of security systems including one to give advance warning of possible satellite collisions. A possible means of construction of an immunocomputer is described in considerable detail, and as with other biologically‐inspired technologies, possibilities of closer association with living systems are explored. An immunocomputer could incorporate real cells and proteins, and might interact with the body as a diagnostic and remedial tool.
It is accepted that much of this is highly speculative at present, but the case for immunocomputing is nevertheless cogently and convincingly argued and there is probably reason to say: “Watch this space”.
Tarankov, A. and Adamatzky, A. (2002), “Virtual clothing in hybrid cellular automata”, Kybernetes, Vol. 31 Nos 7/8, pp. 1059‐72.