peptides [Counts, et al., 2015; Zhou, et al., 2020].
duction
ear algorithms enjoy an excellent interpretation and explanation
y, but may not be sufficient to explain the interplay between
in protease cleavage data or posttranslational modification data.
the decision-tree algorithms and the random forest algorithms
sed to deliver intelligent rules, these algorithms may not deliver
ules because both types of algorithms derive rules based on the
principles. Discovering the best solution for a system among a
mber of candidates is prohibitive and problematic if the exhaustive
has to be used. Avoiding the expensive and exhaustive search
, the use of an optimisation process is considered as a reasonable
Inspired by the Darwin’s natural evolutionary theory, various
ary computation algorithms have been developed, aiming to
natural evolutionary process using a computer program in a
so as to examine how the rule of survival of the fittest can be
or the optimisation problems of rule discovery [Fogel, et al., 1966;
1975; Goldberg, 1989].
are mainly three types of evolutionary computation approaches.
the evolutionary algorithms (EA), the genetic algorithms (GA)
genetic programming algorithm (GP). The evolutionary
ms are normally used to optimise a numerical system with many
rs which are required for optimisation. For instance, the
ary algorithm can be used to optimise the parameters of a neural
This is because the conventional Newton’s method used to
the model parameters is very likely stuck in a local optimum and
tive function of a neural network model is complex [Angeline,
994; Brusic, et al., 1998; Bird, et al., 2019]. The use of the
ary algorithm can start with a pool of initial sets of weights. The
e Newton approach for each of an individual set of weights can
one of many local optimums including the one which is the
ptimum. Among many candidate solutions, the evolutionary