Buy the Book

  GEP Biblio

  Visit Gepsoft


C. FERREIRA, 2002 (Terms of Use) ISBN: 9729589054

Gene Expression Programming: Mathematical Modeling by an Artificial Intelligence

As I said earlier, in the global picture of information metabolism, RNA might be seen as a working copy of a particular sequence of DNA. When a protein gene is expressed, a copy of the gene in the form of messenger RNA (mRNA) is made and used to direct the synthesis of the protein. Thus, in terms of information, the RNA copy contains exactly the same information as the original DNA.

The messenger RNA is not the only kind of RNA molecule working in a cell. Despite its central role in information decoding, structurally, mRNAs are very dull. The structural diversity discussed below is typical of the other classes of RNA, namely transfer RNA (tRNA) and ribosomal RNA (rRNA).

Like DNA, RNA molecules are also long, linear strings of four nucleotides (ribonucleotides, in this case: A, U, C, and G). In contrast to DNA, RNA molecules are single-stranded and some of them are capable of folding in a unique three-dimensional structure. One of the reasons for the folding of RNA molecules resides in the existence of short sequences which are complementary to other sequences within the same molecule. Obviously, if these complementary sequences were to stumble upon each other, short double helixes would be formed. These intramolecular double helixes are indeed fundamental for the unique three-dimensional structure of some RNA molecules.

Thus, like proteins, some RNA molecules can have a unique three-dimensional structure (tertiary structure) and therefore can exhibit some degree of structural and functional diversity. The rules of complementarity in RNA double helixes are much the same as in DNA, with A pairing with U, and C with G. In RNA molecules with tertiary structure, some nucleotides are involved in helix formation and therefore are not chemically available, but other functional groups are free and exposed and thus are free to engage in different kinds of interactions and even participate in biological catalysis. Indeed, this, plus a unique three-dimensional structure, allows RNA molecules to function as real biological catalysts (ribozymes).

So, despite its reduced chemical vocabulary, RNA is the kind of molecule that can simultaneously function as genotype and phenotype, that is, as a simple replicator. Note, however, how in such cases the genotype and the phenotype are tied up together: any modification on the replicator is immediately reflected in its performance. There is no room for subtle or neutral changes in these systems. And subtle and neutral changes are fundamental to an efficient evolution (Kimura 1983; see also section 7.4).

Another important constraint of these simple replicator systems can be very well illustrated using the artificial system of GP, also a simple replicator system. In GP, most modifications made on parse trees result in invalid structures. Indeed, only a very limited number of modifications can be made on GP parse trees in order to guarantee the creation of valid structures. The problem with this system is that extremely efficient search operators such as point mutation cannot be used. Instead, an inefficient sub-tree swapping is used so that valid parse trees are always formed. Nevertheless, no matter how carefully genetic operators are implemented, there are obviously limits to what grafting and pruning can do, and the search space in such systems can never be thoroughly explored.

Home | Contents | Previous | Next