Evolving Protein Motifs Using a Stochastic Regular Language with Codon-Level Probabilities

Experiments involving the evolution of protein motifs using genetic programming are presented. The motifs use a stochastic regular expression language that uses codon-level probabilities within conserved sets (masks). Experiments compared basic genetic programming with Lamarckian evolution, as well as the use of “natural” probability distributions for masks obtained from the sequence database. It was found that Lamarckian evolution was detrimental to the probability performance of motifs. A comparison of evolved and natural mask probability schemes is inconclusive, since these strategies produce incompatible characterisations of motif fitness as used by the genetic programming system.