2014 16th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing
Central Dogma of Molecular Biology- New
Paradigm in Evolutionary Computation
Corina Rotar
Computer Science Department
1 Decembrie 1918 University
Alba Iulia, Romania
crotar@uab.ro
Abstract The aim of this study is to develop a new
evolutionary computation paradigm in terms of molecular
biology. Standard genetic algorithms are heuristics inspired
by the simplified model of natural evolution and genetics.
The latest discoveries and innovations from molecular
biology, related to the conventional central dogma of
molecular biology, generate the necessity of updating the
genetic algorithms, although successfully applied in various
complex tasks. In this direction, the research in Evolutionary
Computation requires a reconsideration of the concepts and
theories underlying the development of these popular
optimization techniques. Since the emergence of the new
features is important in the evolution, the DNA code requires
progress. Evolutionary Computation which is based on the
mutation and the natural selection can be reconsidered in
terms of protein synthesis and reverse transcription. From
the computational perspective, a biological phenomenon
might be interpreted in various forms in order to obtain
reliable computational techniques.
Keywords evolutionary computation paradigm; protein
synthesis; central dogma of biology; optimization
I. INTRODUCTION
In the light of the discoveries and the hypotheses from
biology, related to the popular central dogma of molecular
biology, genetic algorithms, although successfully applied
in various difficult problems seem to be outdated. In this
direction, the research in Evolutionary Computation
requires a reconsideration of the concepts, processes and
theories underlying the development of these popular
optimization techniques. Obviously, the imitation of
biological phenomena requires transformations in order to
obtain effective computational techniques. These required
adjustments are generated by the very nature of the
elements being operated: so in molecular biology and
genetics, the chemical compounds and relationships
through which they relate are fundamental issues, as
computer techniques work with numerically coded
possible solutions which are precisely controlled by
computational tools. By exploiting the specific structures,
processes and concepts which are inspired from biology,
specifically the central dogma of molecular biology - the
optimization problems are reformulated for determining
the DNA code that produces the most efficient protein or
proteome[7] (entire set of proteins expressed by the
genetic material of an organism) consistent with the
problems objectives.
978-1-4799-8448-0/15 $31.00 2015 IEEE
DOI 10.1109/SYNASC.2014.46
Protein synthesis represents the fundamental biological
process by which the cells build their specific proteins. The
central dogma of biology has made simple formula of gene
flow from DNA to protein in the foreground. However, the
modern hypothesis establishes other ways of genetic
transfer: from RNA to DNA by a process called reverse
transcription. This mechanism is interesting in our study,
from the perspective of developing a potential new source
of inspiration in Evolutionary Computation and
optimization tools by which the solutions are created
through an iterative process that improves the outcomes
until reaching the desired accuracy. Since the aim of
developing new biologically inspired computational
technique does not mainly involves the strict copying of
the natural phenomenon, but rather the obtaining of the
best solutions of the problem, certain adjustments of the
original paradigm are justified.
II. FROM NATURE TO COMPUTATION
In the generous landscape of nature-inspired
computation, a computational model that is designed on
the central dogma of molecular biology is ignored. In [1], a
modern perspective is suggested, according to which
nature-inspired computational tools needs revisions in both
theory and applications. Moreover, some criticisms of the
simplified model underlying the standard genetic
algorithm were mentioned in [2]. The algorithms which are
comprised into the Evolutionary Computation are
criticized for their lack of ability to solve other
optimization problem than those which they were designed
for [3]. The multi-criteria, dynamic or multimodal
optimization problems require major adjustments of the
evolutionary algorithm in order to generate the accurate
solutions. The preceding statements are intended to be
arguments to a new perspective on Evolutionary
Computation, that results in a general computational tool
inspired by the central dogma of molecular biology.
A detailed analysis of Natural Computation, given in
[20], defines it as the combination of three major research
directions: 1. computing inspired by nature, 2. simulation
and emulation of nature in computers and 3. computing
with natural materials. Among the listed directions,
computing inspired by nature (e.g. evolutionary
computation, neural networks, artificial immune systems,
swarm intelligence) seems to be the most popular, offering
many computational methods successfully applied in
284
285
 various problems. The proposed approach fits the first
category and focuses on the biological model covered by
the Central Dogma of Molecular Biology, trying to
identify the features that can become sources of inspiration
for designing suitable optimization techniques. The
considered biological paradigm represents in the nature-
inspired computing landscape an interesting and fertile
source of inspiration and in the following paragraphs we
illustrate how a protein synthesis - inspired algorithm can
be extended with minimal effort to solve the various
optimization problems (single and multi-objective).
Recent advances in the field of molecular biology have
attracted the interest of the Natural Computing researchers.
This rich source of inspiration was spotted and there were
developed several methods such as DNA Computing [23]
and Gene Expression Programming[24]. DNA Computing
makes use of the DNA molecules to solve complex
problems. Gene expression programming, basically a
hybrid of the genetic algorithm and the genetic
programming, is defined as an evolutionary technique that
generates computer programs. Unlike the mentioned
approaches, our research focuses on presenting the protein
synthesis metaphor as a source of inspiration for the
computational techniques for solving numerical
optimization problems. Moreover, we design a
straightforward method which can be easily extended for
various types of optimization problems.
III. PROTEIN SYNTHESIS- FROM BIOLOGICAL PARADIGM TO
COMPUTATIONAL PROCEDURE
The following paragraphs will describe several
concepts from the molecular biologys background, based
on which we may develop computational techniques for
solving optimization problems.
Protein synthesis is one of the most fundamental
biological processes by which individual cells construct
their specific proteins. The central dogma of molecular
biology [4],[6] has made simple formula of gene flow from
DNA to protein in the foreground. Later discoveries [5]
completed the original theory by bringing up the reverse
transcription from RNA to DNA, which contradicts the
simple interpretation that genetic information flows only
from DNA to RNA to Protein. This double way transfer
between DNA and RNA represents a fertile source of
inspiration. Therefore, since the central dogma of the
molecular biology is questionable concerning its
incompleteness regarding the backwards genetic transfer
from RNA to DNA (e.g. Retroviruses), the artificial
pattern exploits a specific module of reverse transcription
in order to evolve the DNA code.
A. Central dogma of molecular biology
The central dogma [4] states that the transfer of
biological information is mostly done in the following
direction: DNA can be copied to DNA (DNA replication),
DNA information can be copied into RNA (transcription),
and proteins can be synthesized using the information in
RNA as a template (translation).
286
285
Fig. 1. Graphical description of te DNA, RNA, Protein and genetic
transfer
B. Protein - meaning, function
Protein is a large biological molecule consisting of
chains of amino acids. Proteins perform a vast array of
functions within living organisms, control all cell reactions
and they are responsible for the physical appearances
(phenotype). Genotype, which represents the set of genetic
information that defines an organism, controls phenotype
because genes direct the production of proteins. Proteins,
in turn, dictate every reaction in the cell and therefore are
directly responsible for observable characteristics
(Phenotype). The synthesis of proteins is one of the most
surprising phenomena observed in the cell. The complex
processes by which the genetic information is transmitted
and converted into particular formula can be summed up in
the rule Genotype -> Phenotype. The entire set of proteins
expressed within a cell, tissue, or organism at a certain
time, under defined conditions is called proteome [7]. The
name of the concept derived from the words protein and
genome. Protein synthesis occurs at the level of every cell
in the body in two distinct phases: transcription and
translation (Fig.1).
C. Transcription. From DNA to RNA
The DNA in the genome does not perform protein
synthesis itself, but uses RNA as an intermediary. If the
cell needs a particular protein, a sequence of the long DNA
molecule will first be copied into RNA (transcription).
Next, these RNA copies of DNA segments will be used as
templates to direct the synthesis of the protein
(translation). The flow of genetic information in the cells is
from DNA to RNA to protein. Consequently, each cell
expresses its genetic information according to a principle
so fundamental that it is named the central dogma of
molecular biology.
Single strands RNA molecules produced by
transcription are released from the DNA template. In
addition, because they are copied from only a limited
sequence of the DNA, RNA molecules are shorter than
DNA molecules. Unlike DNA, RNA molecule does not
permanently store the genetic information in cells. In
transcription process, there is a less significant copying
error than that in DNA replication. Replication, and to a
lesser extent, transcription, have been shown to affect the
mutation rates in a variety of genomes: transcription can
be a major source of mutations in nondividing cells. [8].
Recent studies have shown that transcription is more than a
plain mechanism of copying the gene. In [9] there are
described several ways in which the transcription
endangers genome integrity, emphasizing that transcription
is a source of genomic instability. Therefore, transcription
may be exposed as a potential cause of DNA variation.
D. Translation: From RNA to PROTEIN
Translation is the conversion of information from RNA
to information required in the process of building the
proteins. During the translation, the transcribed RNA from
the DNA is decoded by ribosomes for protein synthesis.
RNA leaves the cell nucleus and passes into the cell
cytoplasm. Once there, the "protein factory" of the cell, the
ribosome, reads the message encoded as RNA template
and produces the proteins according to the pattern. The
ribosomes (Palade, 1974) are large and complex molecular
machines of the cells having the main function in the
translation process. During the process of translation,
ribosomes link amino acids together in a specific order as
the RNA molecules indicate. By the contribution of the
ribosomes, the genetic message can be "read" and
therefore the proteins are synthesized.
E. Gene Regulation. Transcriptional Factors
The DNA chain stores the complete information for the
synthesis of all proteins which are required in an organism.
In the process of transcription, RNA copies a fragment of
the genes expressed in the DNA code, which encodes a
specific protein. This phenomenon, called gene expression
is supported by a suite of factors that facilitate copying or
ignoring certain genes. Essentially, gene regulation
represents the natural process of switching the genes on or
off. Gene regulation occurs during transcription phase and
involves the activation of specific proteins according to
different signals received from the environment. These
proteins are called transcription factors and they are
required to initiate or to regulate the transcription process.
Transcriptional factors bind to regulatory regions of a gene
and control the genetic flow from DNA to RNA.
F. Reverse Transcription
Nowadays, it is proven that some viruses [10], whose
genetic material is an RNA molecule, are able to
determine, at the cell level, the synthesis of DNA with
which to replicate. This phenomenon is demonstrated and
it is called reverse transcription [11], [12]. The reverse
transcription is discussed from an evolutionary viewpoint
rather than expressing it from a biological perspective in
[13]. It is stated that reverse transcription represents a
unique tool for transmitting information from the dynamic
RNA to the more inert DNA and has been instrumental
in shaping extant genomes [13]. Virus as a computational
concept is exploited in various fields. For the purpose of
the present work, the virus, the retrovirus more precisely,
represents the special tool through which the reverse
transcription from RNA to the DNA is possible. Using the
mechanism of reverse transcription in the bio-inspired
procedure, it will allow the DNA to evolve during the
search process. In the absence of the reverse transcription,
the system would be less dynamic and would not produce
a rapid convergence towards solutions of the problem
addressed.
In a simple definition, reverse transcription represents
the process by which DNA is synthesized from an RNA
template. Therefore, the genetic information could be
transported in a reverse way, from RNA to DNA, contrary
to the Central Dogma of the molecular biology.
287
286
IV. ARTIFICIAL PROTEIN SYNTHESIS
DNA changes due to mutations. At the macroscopic
level, in evolution, mutation frequency is relatively low.
From the perspective of evolutionary computation,
mutations are produced with a high frequency in order to
obtain the necessary diversity in the search process.
Mutations and natural selection are those processes that
generate evolution. This extremely popular theory is not
entirely proven and the driving force behind evolution is
not fully understood. Besides mutations that are produced
during evolution, DNA can change, for example, by
reverse transcription. Nevertheless, the natural evolution
occurs in a long time. In order to solve the optimization
problems, the basic imitation of the natural evolutionary
processes becomes inefficient. Thus, for defining the
evolutionary algorithms, there are constantly involved
particular strategies such as the limitation of the exploited
resource (small finite populations, limited number of the
generations), the exaggeration of the natural evolutions
features (e.g. a much higher mutation rate), the selection
procedures which favor the best individuals, the crossover
operators that accelerate the convergence, or the
hybridization with traditional computational techniques.
In the same manner as the evolutionary algorithms
include extra ingenious mechanisms without too much
about the biological authenticity, in order to accelerate the
convergence and obtain a better approximation of the
solutions, the proposed optimization method which is
inspired by the central dogma exploits an artificial
procedure such as DNA shuffling.
A. DNA shuffling
Recent advances in biology made possible the
reconstruction in laboratory of DNA-RNA-Protein path.
For objective reasons, in order to rapidly obtain some
useful structures or the desired characteristics, the
artificially driven process, in vitro or in vivo, is
supplemented by certain mechanisms that accelerate the
evolution towards the desired goal. These techniques are
framed into the powerful engineering tool, namely
Directed Evolution [19]. In short, directed evolution can be
defined as the evolving of the proteins toward a user-
defined goal and it is an iterative process which involves
the generation of a set of biological entities of interest and
the screening/selection to identify those variants which
display better properties. The best mutants of the each
iteration will serve as templates for subsequent iterations
of diversification and selection. The process is repeated
until the desired improvement is achieved.
It is surprising to see that, basically, evolutionary
computations metaphor is more comparable to the
Directed Evolution pattern. Considering either the artificial
and the natural proteins synthesiss processes or the
process of evolving a population of possible individuals,
the acceleration of the evolution is desired. Therefore,
diversification process, especially DNA shuffling [14]
proves itself attractive.
Directed Evolution requires a mechanism for
introducing genetic variations. DNA shuffling represents a
method of recombining homologous genes and provides
diversity. The strength of iterative homologous
recombination in computational simulation is mentioned in
[14]. For these considerations, a DNA shuffling-inspired
computational procedure is proposed for producing a better
diversity.
B. Protein synthesis process versus optimization
In the above description, the central dogma of
molecular biology, by its structures and inner processes,
represents an unusually complex biological model that
requires more attention and prefigures a valuable paradigm
in Evolutionary Computation. The biological model
contains several features which must be emphasized:
1. DNA is the genetic signature of an organism. The
information that is stored in DNA is partially transferred
into proteins using RNA as an intermediary. The same
DNA code is responsible for more outputs (proteins). From
these considerations, the DNA structure would be
analogous to the set of possible solutions from the search
space.
2. The observable characteristics of an organism
(phenotype) are generated through the process of genes
expression in two stages: transcription and translation.
The protein synthesis procedure is comparable to the
process of mapping the search space into the objective
space.
3. Proteins guide the reactions in each cell being
directly responsible for the phenotype; therefore, proteins
are the fundamental factors in phenotype. From the
computational point of view, depending by the
optimization problem category, the complete proteome or
only a specific protein represents the required output of the
procedure. The proteins structure corresponds to the
solution in the objective space.
4. The transcriptional factors influence the transcription
process by which DNA is copied into RNA. RNA is not a
clone of the DNA. The process of transcription is
responsible for gene expression, respectively, for the
expressed output of the protein synthesis process. For the
bio-inspired computational technique, the transcriptional
factors would represent the probabilities of transcriptional
errors during the transcription process.
5. By reverse transcription, the information is
transmitted backwards, from RNA to DNA, contrary to the
central dogma of molecular biology. In developing the
optimization technique, the reverse transcription procedure
infuses new promising zones of the search space into the
DNA. Further, the protein synthesis procedure creates
improved outputs according to the problems objectives.
6. DNA shuffling is an artificial technique that is
designed in order to achieve a diverse genetic pool in
Directed Evolution; it could be appropriated for our
purpose as it is a process by which the genetic code is
altered and therefore the genetic diversity improves.
V. PROTEIN SYNTHESIS ALGORITM FOR
OPTIMIZATION
Protein synthesis algorithm (PSA) is an approach that
is inspired by the combination of two important processes:
the biological paradigm of protein synthesis and the DNA
288
287
shuffling technique which is framed in the Directed
Evolution. The method is valuable as it offers the
possibility to address a large range of optimization
problems.
General pattern for re-interpreting an optimization
process as a protein synthesis mechanism consists in
following steps:
0. DNA shuffling - the process by which the DNA code
varies; it is an artificial process that is designed in order to
achieve a diverse genetic pool.
1. Transcription the process by which the genetic
code is transcribed into RNA code
2. Translation the process of producing the proteins
from RNA code
3. Regulation the process which controls the
transcription
4. Reverse transcription the process by which RNA
code is reverse transcribed in DNA genetic code
A. PSA for single objective optimization
In the case of single objective optimization (SOP), the
goal is to determine the best possible solutions from the
search space. From this point of view, the biological
paradigm described in the previous paragraphs can be
comprehended as the iterative process for the synthesis of
a protein with the desired features from a diverse library of
genes. The DNA code resulting after the DNA shuffling
process is viewed as a set of possible solutions in the
search space. At each iteration, to produce proteins
increasingly skilled according to the given objective, the
following main procedures are exploited: transcription,
translation and reverse transcription.
B. DNA, RNA and PROTEIN for SOP
Let us consider n the size of the search space, and f- the
single objective. A solution from the search space is
represented by a gene of the DNA chain. Each gene
codifies a possible solution: ( x1 , x2 ,, xn ). DNA chain
codifies the solutions in the search space and it is given by
the vector:
DNA={( x11 , x12 , ..., x1n ),( x12 , x22 , ..., xn2 )...( x1l , x2l , ...,
xnl ) }, where l- length of the DNA chain.
The RNA chain has the same size and structure, as being a
transcription of the DNA template:
RNA={( y11 , y12 , ..., y1n ) ( y12 , y 22 , ..., y n2 ) ... ( y1l , y 2l ,
..., y nl ) }, where l- length of the RNA chain and each
copied gene corresponds to one specific protein. Each
protein represents the objectives value of a gene,
considering the RNA chain.
Proteins -represent the set of RNA code translated into the
objective space:
Proteins={
( ) ( ) ( )
f y11 , y12 ,..., y 1n , f y12 , y 22 ,..., y n2 ,... f y nl , y nl ,..., y nl }.
C. PSA for multi-objective optimization
In the case of a multi-objective optimization problem
(MOP), the aim is to find the set of Pareto optimal
solutions. In this sense, approaching the multi-criteria
problem in terms of protein synthesis procedure is
performed in the following way: considering DNA (the
genotype) as a set of possible solutions in the search space,
each solution being encoded as a gene, the aim is to design
the proteome (phenotype) that best satisfies the conditions
of the problem. Thus, the requirement is to generate all
expressed proteins which correspond to the best
approximation of the Pareto front.
Fig. 2. Multiobjective optimization. Computational model: Protein
synthesis, structures and processes
D. DNA, RNA and Protein for MOP
Let us consider n the size of the search space. A
solution from the search space is represented by a gene of
the DNA chain. Each gene codifies a possible solution:
( x1 , x2 ,, xn ).
Considering l- the length of the DNA chain, the vector
DNA codifies a set of possible solutions in the search
space and it is represented by the following array:
DNA={( x11 , x12 , ..., x1n ),...( x1l , x2l , ..., xnl ) }.
The output of the PSA algorithm is given by the final
DNA chain that contains an approximation of the Pareto
set. The usual termination condition of PSA is given by
attaining the maximum iterations number and not by
generating a list that contains only Pareto non-dominated
solutions. Therefore, the DNA could comprise residual
dominated solutions which can be further removed.
The RNA chain has the same size and structure, as being a
transcription of the DNA template:
RNA={( y11 , y12 , ..., y1n ),... ( y1l , y 2l , ..., y nl ) }, where l-
length of the RNA chain and each copied gene
corresponds to one specific protein.
The Proteome (totality of expressed proteins) contains an
approximation of the Pareto front considering the DNA
chain and it is given by:
1 1 1 l l l
Proteome= {( f , f ,... f ), ( f , f ... f )}, where m-
1 2 m 1 2 m
represents the number of objectives.
E. Protein synthesis as a general optimization tool
In single objective optimization case, the protein
synthesis process is seen as the iterative process of
searching best gene/genes which correspond to the desired
protein and thus maximizes the single objectives value.
For multi-objective optimization, protein synthesis
process corresponds to the procedure by which an
approximation of the Pareto optimal set converts the
289
288
genetic material (information) into the approximation of
the Pareto front (proteome). The core of Protein synthesis
mechanism consists of three stages, as in the
corresponding biological paradigm. Firstly, the
transcription process takes part and involves the copy of
the genetic material (DNA chain) and also implicates a
series of regulated artificial small alteration of the DNA
chain in order to create the RNA templates. Secondly,
RNA is transformed into proteins through the translation
mechanism. This mechanism is simple due to the
straightforward translation of the solution from the search
space into the objective space in multi-objective
optimization situation. Thirdly, the reverse transcription
inserts variation into the original DNA chain. As the
genetic material needs to evolve, and reverse transcription
is not powerful enough to give the desired speed of the
evolution, an artificial procedure is taken into account,
namely DNA shuffling.
F. DNA Shuffling
DNA shuffling [14] represents a widely used method in
protein engineering technology for creating a diverse gene
library and it is one of the most representative processes
of diversification in Directed Evolution.
Since the purpose of computational techniques is to
determine the solutions of the problem in a relatively good
time, the DNA shuffling process becomes an idea that is
worthy of consideration. The core of the protein synthesis,
which is inspired by the central dogma (DNA-> RNA>
Protein) is enclosed in an iterative process involving an
acceleration of DNA evolution by applying the DNA
shuffling process. Thus, each gene from the DNA chain
blends with its homologous gene. Homologous gene
represents the closest different gene in the search space.
For simplicity, we use the Euclidian distance as a measure
of similarity.
The resulted gene will survive among the DNA chain
if it is superior to the original gene. The performance of
genes is measured differently depending on the type of
problem considered. When referring to maximization
problem with a single objective, the comparison of parent-
descendent genes is based on the values of the objective
function. For the optimization problems that involve two
or more objectives, the definition of Pareto dominance is
considered. Therefore, the operator > from the next
procedure has different meanings depending on the
problems type.
Procedure DNA_Shuffling:
Input: DNA={( x11 , x12 ,..., x1n ),...,( x1l , x2l ,..., xnl )}
For each i=1 to l
Select the jth gene that is homologous of ith gene
( y1i , y 2i ,..., y ni )=Shuffling(( x1i , x2i ,..., xni ),( x1j , x2j ,..., xnj )
)
If ( y1i , y 2i ,..., y ni ) > ( x1i , x2i ,..., xni ) then
( x1i , x2i ,..., xni )=( y1i , y 2i ,..., y ni )
Output: DNA={( x11 , x12 ,..., x1n ),...,( x1l , x2l ,..., xnl )}
EndProcedure
 Shuffling procedure corresponds to the crossover
operator in GA paradigm. For our study we used the
discrete crossover as it is descibed in [21].
G. Transcription
Transcription represents the process by which the
RNA chain is created. In this procedure, the DNA chain is
the basis for the RNA chain. Each gene from DNA has its
own probability to be copied as it is or altered before it is
passed to the RNA. The probabilities to alter the genes are
given by the transcriptional factors which are computed
with respect to the quality of the genetic sequence.
Considering TF={ 1 ,, l } the array of transcriptional
factors, the RNA is synthesized by the following
algorithm:
Procedure Transcription:
Input:TF={ 1 , 2 ,, l }
DNA={( x11 , x12 ,..., x1n ),...,( x1l , x2l ,..., xnl )}
For each i=1 to l
For each j=1 to n
If rnd < i then
( )
x ij + rnd max j x ij , if rnd 0.5
i
( )
x j rnd x j min j , otherwise
i
i i
Else y = x
j j
Output: RNA={( y11 , y12 ,..., y1n ),...,( y1l , y 2l ,..., y nl )}
EndProcedure
The values min j and max j represents the limits of
the search interval for the jth variable.
H. Translation
Translation represents the protein synthesis process.
By using the RNA chain obtained during the transcription
procedure, translation constructs the protein or the set of
proteins (the proteome) by computing the objective values
for the solutions codified in the RNA.
Procedure Translation:
Input: RNA={ ( y11 , y12 , ..., y1n ), ..., ( y1l , y 2l , ..., y nl ) }
For each i=1 to l
For each j = 1 to m /*(m is 1 for single objective problems)*/
f ji = Compute jth objective of ith solution ( y1i , y 2i ,..., y ni )
Output:
Proteins/Proteome= ( f11 , f 21 ,..., f m1 ),,( f1l , f 2l ,..., f ml )
EndProcedure
I. Regulation. Transcriptional factors
From the description of biological model we
emphasize the main point that the transcriptional factors
represent indispensable elements in the transfer of
information from DNA into RNA. In simple words, their
role is defined by the effect they have on copying the
genetic information from DNA in the intermediate RNA.
By ignoring the intricate biological details about the
transcriptional factors action we develop a computational
procedure where these elements are reinterpreted as
extents of gene expression in the process of transcription.
For these considerations, transcriptional factors (TF)
could represent activation or inhibition thresholds in gene
expression during transcription from DNA to RNA.
Although the biological model is more strict about
transcriptional factors action, either as activators by
triggering the gene transcription, or as inhibitors by
causing the omitting of some sequences from DNA, the
computational model mimics the action of TFs by using
continuous values that affect the accuracy of copy
(accuracy in transcription) rather than the gene occurrence
in RNA template.
The transcriptional factors are given by the values TF
which represent the degrees of dissimilarity between the
original sequence of genes from the DNA chain and the
genes in the RNA chain. Thus, a null value indicates that
the original DNA sub-sequence is cloned into the
corresponding location into the RNA. Otherwise, for a
non-zero value, the original genes are altered and then
copied into RNA. These transcriptional errors would
sustain further the diversity in the search space.
Considering l the length of the DNA chain, the
y ij =
transcriptional factors are computed as follows:
i =
(
rank 1
i
)
+ (1 ) i , i {1,..., l} ,  randomly
l l
generated in [0,1], and:
- rank i in multi-objective context, represents the non-
domination rank [15] of the ith gene among the other
genes of the DNA chain, taking into account the given
objectives; for single objective optimization problems, it
corresponds to the rank of the ith gene, taking into account
the single objectives values
- i - represents the number of all genes from the DNA
chain which are in the vicinity of the ith gene; the radius of
the neighborhood  is computed as the average of the
Euclidian distances between each distinct pair of genes
l 1, if euclidian_distance(i, j)
i = g (i, j ) , g (i, j ) =
j 0, otherwise
Consequently, the transcriptional factor is a convex
combination of the rank and the crowding degree.
Regulation procedure computes the transcriptional factors
by using the following measures:
- (in genotype) the crowding of the solutions that are
codified in RNA template as genes
- (in phenotype) the ranks of the solutions in the objective
space that are codified in the corresponding protein.
Therefore, transcriptional factors are affected by the
genotype and phenotype accordingly. Further, these
transcriptional factors are used in the transcription
procedure as transcriptional error probabilities.
Procedure Regulation:
Input: RNA={( y11 , y12 ,..., y1n ),..,( y1l , y 2l ,..., y nl )}
Proteins (Proteome) =( f11 , f 21 ,..., f m1 ),,,( f1l , f 2l ,..., f ml )
For each i=1 to l
290
289
 Compute the transcriptional factor i Non-dominated Sorting Genetic Algorithm II [15] for
multiobjective optimization problems.
Output: TF={ 1 , 2 , , l }
EndProcedure A. Single-objective optimization
J. Reverse transcription DNA synthesis For single-objective optimization, we consider the
In a simple definition, reverse transcription represents following test functions:
the process by which DNA is synthesized from the RNA Schaffers
f1 (x, y ) = 0.5 +
( )
sin 2 x 2 y 2 0.5
(1 + 10 (x + y ))
template. The biological process of reverse transcription F6 3 2 2 2
proves itself as a suitable mechanism for our purpose: as
long as the intention is to create the best possible DNA (x, y ) [ 100,100]2 , Global min: f1* (0,0) = 0.
chain which codifies the proteins/proteome, the DNA
chain should be transformed repeatedly. From this point
Ackley
(
f 2 (x, y ) = 20 exp 0.2 0.5 x 2 + y 2 -

)
of view, the reverse transcription mechanism represents a
exp(0.5(cos(2x ) + cos(2y ))) + 20 + e
proper technique to evolve the artificial DNA chain
corresponding to the problems solutions. Therefore, the (x, y ) [ 30,30]2 ,Global min: f 2* (0,0) = 0.
role of the reverse transcription procedure consists in Rastrigin f 3 (x, y ) = 20 + x 2 10 cos(2x ) + y 2 10 cos(2y )
adjusting the DNA chain in order to generate highly
qualified proteins. Thus, the RNA chain is copied in the (x, y ) [ 5,5]2 , Global min: f 3* (0,0) = 0.
DNA chain. Further, the newly transformed DNA is the The GA, PSO and PSA run for 20 times. We considered
basis for the next protein synthesis. The importance of the following parameters and setups for each run time:
reverse transcription is given by its strength in
transforming the inert DNA chain. By using this PSA 100 generation, Size of DNA=50, Size of RNA=50
procedure, better solutions are inserted backward into the GA 100 chromosomes, 100 iterations
DNA chain and contribute to the evolution. standard convex crossover (pc=0.9)
random uniform mutation (pm =0.1)
Procedure ReverseTranscription: fitness proportional selection
Input: RNA={ ( y11 , y12 , ..., y1n ),..., ( y1l , y 2l , ..., y nl ) } PSO 100 particles, 100 iterations
inertia weight linearly decreases from 0.9 to 0.4
For each i=1 to l
learning coefficients = 2.0, maximum velocity =10
If( y1i , y 2i ,..., y ni )>( x1i , x2i ,..., xni ) then
TABLE I. PSA VERSUS GA, PSO. COMPARISON RESULTS.
( x1i , x2i ,..., xni )=( y1i , y 2i ,..., y ni )
Test
Output: DNA={( x11 , x12 ,..., x1n ),...,( x1l , x2l ,..., xnl )} function
Schaffer Ackley Rastrigin
EndProcedure Mean 9.3E-16 8.3E-03 1.55E-13
PSA Stddev 7.8E-16 2.8E-03 2.5E-13
Depending of the problems type, single or multi-
objective optimization, the comparison between the Minimum 2.22E-16 8.28E-03 1.42E-14
sequence of RNA and original DNA is made either Mean 2.55E-03 1.18E-03 3.83E-04
according to the Pareto dominance or according to the PSO Stddev 4.14E-03 9.05E-04 8.17E-04
values of the single objective. Minimum 4.21E-07 1.07E-04 3.62E-06
Mean 6.72E-04 2.94E-02 2.78E-04
Next the outline of the PSA algorithm is given: Stddev 5.46E-04 2.66E-02 4.66E-04
GA
Algorithm PSA is Minimum 1.59E-04 9.22E-04 2.51E-07
DNA initialization
Loop Table I presents the average values of the best proteins
DNA_Shuffling found by each algorithm. The results show that PSA
TRANSCRIPTION performs better than standard GA for the considered test
functions. Also, for the Ackley test function, PSO offers
TRANSLATION
better results than both PSA and GA.
REGULATION
REVERSE_TRANSCRIPTION B. Multiobjective optimization
Until StoppingCondition is True In order to investigate the performance of PSA, we
End. used several well-known test problems: ZDT1, ZDT2,
ZDT3 [16] as the two objectives test functions with 30
VI. RESULTS variables, and DTLZ1, DTLZ2 and DTLZ3 problems [17]
The proposed algorithm is tested both for single- as the three objectives test functions with 7 variables
objective optimization and multi-objective optimization. (DTLZ1), respectively, 12 variables (DTLZ2 and DTLZ3).
For each type of problem (single or multi-objective), For performance assessment, we compute hyper-volume
several popular test functions are chosen and the metric (HV)[17]. Hyper-volume metric corresponds to the
comparisons are conducted against the popular algorithms: size of the objective space which contains the solutions
Genetic Algorithm and Particle Swarm Optimization which are Pareto - dominated by at least one of the
(PSO) [22] for single objective optimization problems and members of the set.

291
290
 Independent-samples t-tests were conducted to REFERENCES
compare PSA results and NSGA2 results for ZDT and [1] Banzhaf, W., et al. "Guidelines: From artificial evolution to
DTLZ test functions. There were significant differences in computational evolution: a research agenda." Nature Reviews
the hypervolume metric for PSA and NSGA2, considering Genetics 7.9 (2006): 729-735..
99% confidence intervals. The results summarized in [2] Donald S. Burke, Kenneth A. De Jong, John J. Grefenstette,
Table II shows that PSA perform better than NSGA2 for Connie Loggia Ramsey, and Annie S. Wu. 1998. Putting more
genetics into genetic algorithms. Evol. Comput. 6, 4 (December
the all considered problems, except DTLZ2. 1998), 387-410.
TABLE II. PSA VERSUS NSGA2: STATISTICAL RESULTS FOR HV [3] Wolpert, D.H., Macready, W.G. (1997), "No Free Lunch Theorems
for Optimization", IEEE Transactions on Evolutionary
Test HV (PSA) HV (NSGA2) Computation 1, 67.
T_Stat p
function Mean (SD) Mean (SD) [4] Crick, F (August 1970). "Central dogma of molecular biology.",
ZDT1 8.71E-01 7.77E-01 5.38 1.12E-04 Nature 227 (5258): 5613.
4.57E-02 2.02E-02 [5] Temin H, Baltimore D. RNA-Directed DNA Synthesis and RNA
ZDT2 7.80E-01 5.08E-01 1.12E+01 1.E-06 Tumor Viruses. In Advances in Virus Research, vol. 17. Smith
7.36E-03 7.50E-02 KM, Lauffer MA, and Bang FB, Eds. New York: Academic Press,
ZDT3 6.52E-01 5.78E-01 9.06 8.82E-06 1972, pp. 129-186
1.44E-03 2.40E-02 [6] Watson JD, Crick FH. Molecular structure of nucleic acids; a
DTLZ1 9.38E-01 7.14E-01 4.75 3.90E-04 structure for deoxyribose nucleic acid. Nature. 1953 Apr
7.22E-04 2.15E-02 25;171(4356):737738
DTLZ2 8.84E-01 8.90E-01 -4.05 3.44E-04
[7] Wilkins, M. R. et al., From proteins to proteomes: large scale
1.17E-05 1.24E-05 protein identification by two-dimensional electrophoresis and
DTLZ3 2.69E-01 1.05E-01 8.08 1.59E-07 amino acid analysis. BioTechnology14, 6165 (1996).
2.51E-03 1.61E-03
[8] Polak P, Arndt PF. Transcription induces strand-specific mutations
at the 5 end of human genes. Genome Res 2008;18:1216-1223.
VII. CONCLUSIONS
[9] Kim N, Jinks-Robertson S (2012) Transcription as a source of
The flow of the concepts and ideas propagates in both genome instability. Nat Rev Genet 13: 204214.
ways, from nature toward computation and vice versa so [10] Robin A Weiss, The discovery of endogenous retroviruses,
that, it can be said that the artificial and the natural Retrovirology, 2006, Volume 3, Number 1, Page 67.
contexts merge harmoniously. The field of nature-inspired [11] Temin HM, Mizutani S (June 1970). "RNA-dependent DNA
computation grows rapidly and includes numerous bio- polymerase in virions of Rous sarcoma virus". Nature 226 (5252):
12113.
inspired paradigms that serve as practical models for
[12] Baltimore D (June 1970). "RNA-dependent DNA polymerase in
various techniques. Besides, novel discoveries in virions of RNA tumour viruses". Nature 226 (5252): 120911.
molecular biology occur with greater rapidity, making the [13] Mourie: Reverse transcription in genome evolution. Cytogenet
two directions to appear unsynchronized. For these Genome Res. 2005; 110(1-4):56-62.
reasons, a pattern that is inspired by the latest discoveries [14] Stemmer WP. Rapid evolution of a protein in vitro by DNA
and artificial technologies, which are framed in genetics shuffling. Nature, 1994 Aug 4;370(6488):389-91
and molecular biology, becomes required. In our study, we [15] Deb, K., Pratap, A., Agarwal, S., Meyarivan, T. (2002a), A fast
emphasize the richness of such a model and we propose and elitist multi-objective genetic algorithm: NSGA-II, IEEE
the paradigm of protein synthesis as a new perspective in Transactions on Evolutionary Computation, 6 (2), pp. 182-197.
the Evolutionary Computations landscape. [16] Zitzler, E., Thiele, L., Laumanns, M., Fonseca, C. M., & Da
Fonseca, V. G. (2003). Performance assessment of multiobjective
The algorithm that is inspired by the natural process of optimizers: An analysis and review. Evolutionary Computation,
protein synthesis and by the artificial DNA shuffling is IEEE Transactions on, 7(2), 117-132.
expected as a general optimization technique, by offering a [17] Zitzler, E., Deb, K., Thiele, L. (2000), Comparison of
simple use for different problems. This paper presents Multiobjective Evolutionary Algorithms: Empirical Results,
Evolutionary Computation, vol. 8 no, 2, pp. 173-195.
experimental preliminary results obtained with PSA for the
[18] Deb, K., Thiele, L., Laumanns, M., Zitzler, E. (2002b), Scalable
optimization problems with one or two objectives, taking Multi-Objective Optimization Test Problems, in CEC 2002:
into account various problems and comparing the Proceedings of the conference on Congress on Evolutionary
performance of the proposed approach with the popular Computation, IEEE Press, pp. 825 - 830.
standard techniques: GA, PSO and NSGA2. The results [19] Cobb, R. E., Chao, R. and Zhao, H. (2013), Directed evolution:
show that PSA is a suitable optimization technique, being Past, present, and future. AIChE J., 59: 14321440. doi:
able to offer better solutions in most cases. 10.1002/aic.13995.
[20] De Castro, L.N.. Fundamentals of natural computing: basic
The major advantages of the described method consist concepts, algorithms, and applications. CRC Press, 2006.
in both its comparable performance to the existing [21] Beyer, H. G., & Schwefel, H. P. (2002). Evolution strategiesA
algorithms performances, but also with the possibility to comprehensive introduction. Natural computing, 1(1), 3-52.
easily address the different optimization problems. As [22] Shi, Y., & Eberhart, R. (1998, May). A modified particle swarm
further research, addressing the optimization problems in optimizer. In Evolutionary Computation Proceedings, 1998. IEEE
World Congress on Computational Intelligence., The 1998 IEEE
dynamic environment and more numerical experiments are International Conference on (pp. 69-73). IEEE.
required to prove the strength of the proposed paradigm.
[23] Adleman, L. M. (1994). "Molecular computation of solutions to
Our study encourages a deeper exploitation of the protein combinatorial problems". Science 266 (5187): 10211024.
synthesis metaphor by analyzing the complex biological [24] Ferreira, C. (2001). "Gene Expression Programming: A New
structures and processes to design new computational Adaptive Algorithm for Solving Problems". Complex Systems,
techniques. Vol. 13, issue 2: 87129 .

292
291