The notion of derivations corresponding to gene products would appear to be a useful one, since it formally establishes the analogies between parsing and gene expression, and between parse trees and gene structure, which are inherent in the first sample gene grammars given above. It also allows us to adapt the discussion to wider questions of the differential control of gene expression in different tissues and developmental stages.
For example, if we equate successful parsing with gene expression, we must concede that a substring that is a gene at one time may not be a gene at another. This is troublesome, unless we view genes in the context of the genome as a whole. If the genome is inherently ambiguous, then multiple global derivations could correspond to particular cell types at particular times and under particular conditions. Any given derivation may or may not call for the gene sub derivation in question.
From this viewpoint, it might be better to name the corresponding nonterminal expressed-gene rather than simply gene. Does this mean, though, that in any given fixed global state of differentiation, etc., genes and gene expression may yet be deterministic? For, at a local level the apparent ambiguity of overlapping genes, or of expressed vs. unexpressed genes, does not mean that such an ambiguity necessarily exists at any given time in the cell; there may be external, regulatory factors that “tip off” some cellular recognizer and thus specify one or the other of the available parses.
In this model there could be distant elements specifying such regulatory factors in an overall genomic language, acting against ambiguity that may otherwise be present within isolated segments. Indeed, grammar-based approaches have been proposed for simulating gene regulatory systems [Searls, 1988], and for modelling their genomic arrangement using.
However, such mechanisms by and large exert their effects via exogenous elements (such as DNA-binding proteins) whose biochemical activity would seem to be necessarily “ambiguous,” if only at threshold levels.
Conclusion:
It is difficult to imagine a language recognizer sophisticated enough to precisely simulate regulation that ultimately depends on the physical chemistry of molecules moving through a cell. Thus, whatever the cell might do to chart its fate deterministically, would seem to be inaccessible to any linguistic description of the genome out of context.
