Q&A from Peer Reviews

These are some of the questions posed by the excellent and generous (yes, generous, because s/he even went to great lengths to correct my English) reviewer at Complex Systems, whose comments contributed to make my paper a much better one. Many thanks!

• What are the implications for Wolpert and Macready's "No Free Lunch"...
• What is the rationale for describing GEP genes as having a head and a tail?...
• In describing genetic operators as insertion sequences and transposition...
• You say that "duplication of genes plays an important role in biology and...
• I do not understand why mean fitness should fluctuate as in Figures...

What are the implications for Wolpert and Macready's "No Free Lunch" Theorem for the generality of this claim? Should we expect that GEP will always outperform alternative strategies?

The NFL theorem states that over all possible search spaces (smooth, neither strictly smooth nor strictly random, and random) all search methods perform equally well including simple random guess. But fitness landscapes are far from random and in those landscapes some techniques perform better than others. In this paper I show that GEP performs very well over a wide range of applications.
 

What is the rationale for describing GEP genes as having a head and a tail? It makes intuitive sense, but is confusing in light of the genetic terms (e.g., ORFs, genes, chromosomes, etc.) used throughout the narrative. Would some alternative pair of terms be equally descriptive, but more appropriate in this context?

I cannot think of nothing better. Note that similar, colorful designations are used in Molecular Biology (and in Computer Science, for that matter), for instance, the “cap” and the poli(A) tail.
 

In describing genetic operators as insertion sequences and transposition, the use of biological terms is inaccurate. In DNA molecules, an insertion sequence is "a transposable element that carries no genetic information except that which is necessary for transposition" (Li & Graur, 1991, Fundamentals of Molecular Evolution, Sinauer and Associates, Sunderland MA, p. 240; see also Li, 1997, Molecular Evolution, Sinauer and Associates, Sunderland MA). Transposition is "the movement of genetic material from one genomic location to another" (ibid, p. 250), and duplication is "the presence or the creation of two copies of a DNA segment in the genome" (ibid, p. 237).

I give a clear definition of the three kinds of transposable elements used in GEP. Besides, they do in fact have a resemblance with natural transposable elements, namely the capacity of jumping from one place to another and the possibility of getting copied in the process. Compare this choice of terms with the notion of “gene” and “allele” in GAs, or “introns” in GP.
 

You say that "duplication of genes plays an important role in biology and evolution...Interestingly, in GEP, individuals with duplicated genes are commonly found...". I have two complaints about this part of the arguments. First, the observation that duplication occurs in living systems does not help explain why it is important or helpful in GEP. Second, the reason that duplications are found in GEP searches can be simply explained by the existence of a genetic operator for duplication (misnamed gene transposition). The argument is circular. (Gene duplication is important in evolution. We observe duplication in GEP. Thus, duplication is important.)

Duplication of genes in nature not only occurs but is also important: for example, the multigene families of several important proteins (e.g., the globins), the duplicated rRNA genes, etc.

Concerning the second complaint: There is no such operator in GEP: it is gene transposition in conjunction with one of the three kinds of recombination that can give rise to duplicated genes. During gene transposition the transposable gene is deleted at the place of origin. This is clearly explained in section 5. (This kind of misunderstanding is one of the reasons I think the genetic operators should not be removed from the main text.)
 

I do not understand why mean fitness should fluctuate as in Fig. 10 and Fig. 12. Why does mean fitness decrease from one generation to the next? Even after a perfect solution exists in the population, mean population fitness continues to oscillate. Does this suggest that the evolutionary dynamics might be adjusted to search the solution space more effectively?

I included new studies in order to show possible evolutionary dynamics in GEP and to compare them with other dynamics (see Figure 11). (This kind of questioning is also another reason why I think genetic operators should not be removed from the main text.)
 

Questions & Answers

***