(

)

,

p

N chromosome using the min-max rule for a GP modelling is

mplex, ((cH(eFbL)+)+(bT(bQaGcYdG)+)&)&. In this rule, the

ubunit (eFbL) involves two residues, but the fourth subunit

YdG) involves four residues as variables.

A tree structure of a complicated RPN chromosome for the min-max function.

ext question was the fitness of each RPN rule formulated using

max function. Because analysing the factor Xa protease cleavage

pattern discovery was a classification problem, therefore

s (8.8) and (8.9) were used for the fitness analysis.

ool size was 100, and the value of ߙ was 0.9 for the factor Xa

cleavage pattern discovery. The mutation, dual-chromosome

r, and single-chromosome crossover operations were randomly

with an equal probability during a learning process. The learning

was terminated when the top fitness measurement was unchanged

than 20 cycles or maximum learning cycle was approached. Two

matrices were used to measure the similarity between amino

d used for a comparison. One was the Dayhoff mutation matrix

and Schwartz, 1978] and the other was the BLOSUM62

matrix [Henikoff and Henikoff, 1992].

dition to three breeding operators, a special treatment as a

ent to the old approach [Yang, et al., 2003] was also introduced.

ment was used to ensure a sufficient diversity within a population

promote the breeding performance, i.e., generating more diverse

omes to allow new and better ones to be generated through