IEEE Computational Intelligence Magazine - November 2021 - 85

Michail-Antisthenis Tsompanas,
Larry Bull, Andrew Adamatzky
University of the West of England
Application
Notes
Igor Balaz
University of Novi Sad
Evolutionary Algorithms Designing Nanoparticle
Cancer Treatments with Multiple Particle Types
Abstract
T
here is a rich history of evolutionary
algorithms tackling optimization
problems where the
most appropriate size of solutions,
namely the genome length, is unclear a
priori. Here, we investigated the applicability
of this methodology on the problem
of designing a nanoparticle (NP)
based drug delivery system targeting
cancer tumors. Utilizing a treatment
comprised of multiple types of NPs is
expected to be more effective due to the
higher complexity of the treatment. This
paper begins by using the well-known
NK model to explore the effects of fitness
landscape ruggedness on the evolution
of genome length and, hence,
solution complexity. The size of novel
sequences and variations of the methodology
with and without sequence deletion
are also considered. Results show
that whilst landscape ruggedness can
alter the dynamics of the process, it does
not hinder the evolution of genome
length. On the contrary, the expansion
of genome lengths can be encouraged
by the topology of such landscapes.
These findings are then explored within
the aforementioned real-world problem.
Variable sized treatments with multiple
NP types are studied via an agent-based
open source physics-based cell simulator.
We demonstrate that the simultaneous
evolution of multiple types of NPs leads
Digital Object Identifier 10.1109/MCI.2021.3108306
Date of current version: 13 October 2021
to more than 50% reduction in tumor
size. In contrast, evolution of a single
NP type leads to only 7% reduction in
tumor size. We also demonstrate that the
initial stages of evolution are characterized
by a fast increase in solution complexity
(addition of new NP types),
while later phases are characterized by a
slower optimization of the best NP
composition. Finally, the smaller the
number of NP types added per mutation
step, the shorter the length of the
typical solution found.
I Introduction
Evolutionary algorithms (EAs) can be
applied to problems of unknown complexity
through the use of variablelength
genomes. Here the term genome
refers to the set of variables for the given
optimization problem. A seminal example
of the variable-length genome
methodology is the work by Fogel et al.
[1] on finite state machine design
through the use of a mutation-based
scheme that can increase or decrease the
number of nodes. A subset of variablelength
problems, known as metameric
representation problems [2], is tackled by
utilizing a segmented variable-length
genome. This means that the solutions
are defined as sets of similar components.
Examples of these problems
include the layout of wind farms, wireless
sensor networks, and composite
Corresponding author: M.A. Tsompanas, University of the
West of England, (email: Antisthenis.Tsompanas@uwe.ac.uk).
laminate stacking problems [3]. For
example, the optimization of the placement
of wind turbines on a predefined
site with a specific wind profile can
enhance the overall efficiency of the
wind farm by limiting the turbine interactions
[4], [5]. Since there is no given
number of turbines in the problem, a
variable-length representation can be
utilized. Similarly, in coverage problems,
such as the planning of cellular systems
[6] or of wireless sensor networks [7],
[8], the placement of an unknown number
of nodes to achieve the coverage of
a sector is required. Despite the fact that
the amount of nodes is not predefined,
the optimization process has to take into
account minimizing costs whilst maximizing
the reliability and coverage of
the instalment. Variable-length algorithms
proved to be more efficient than
the fixed-length ones in some cases,
even when the optimal amount of components
of the solution was known [2].
Other examples of variable-length representations
include aspects within the
field of electrical circuit design, such as
designing passive filters [9], transistor
amplifiers [10] and computing circuits
[11], and the field of neuroevolution
[12]-[15], where artificial neural network
weights and topology are often
optimized concurrently (after [16]).
This paper first expands on previous
variable-length studies by investigating
how the ruggedness of a fitness landscape
influences genome length during evolution,
with an abstract tunable model and
1556-603X/21©2021IEEE
NOVEMBER 2021 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE 85
IEEE Computational Intelligence Magazine - November 2021

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2021
