=Paper=
{{Paper
|id=Vol-2098/paper30
|storemode=property
|title=Construction of Optimal Immune Network Model Based
on Swarm Intelligence Algorithms for Computer-aided
Design of New Drugs
|pdfUrl=https://ceur-ws.org/Vol-2098/paper30.pdf
|volume=Vol-2098
|authors=Galina A. Samigulina,Zhazira A. Massimkanova
}}
==Construction of Optimal Immune Network Model Based
on Swarm Intelligence Algorithms for Computer-aided
Design of New Drugs ==
https://ceur-ws.org/Vol-2098/paper30.pdf
Construction of Optimal Immune Network Model Based
on Swarm Intelligence Algorithms for Computer-aided
Design of New Drugs
Galina A. Samigulina and Zhazira A. Massimkanova
Institute of Information and Computational Technologies, Almaty, Kazakhstan
galinasamigulina@mail.ru
masimkanovazh@gmail.com
Abstract. Nowadays the development of information technologies based on
bioinspired intellectual approaches for the computer design of new drugs and
forecasting of their properties is an urgent task. The research is devoted to the
development of an intellectual information system for conducting scientific re-
searches and forecasting the structure-property/activity relationship of new
drugs based on artificial immune systems approach. In accordance with the
concept of multi-algorithmic approach, the construction of an optimal immune
network model and the allocation of informative descripts are carried out using
swarm intelligence algorithms: modified algorithms of ant colonies and particle
swarms. The developed information system allows selecting the best algorithm
for preliminary data processing, in which after immune network modeling, the
value of generalization error will be minimal. The use of multi-algorithm ap-
proach at immune network modeling of drugs requires the systematization of
used algorithms and the creation of an integrated ontological model, which al-
lows structuring the input and outputting data. There is presented an example of
the database of sulfanilamides with different pharmacological activity, also
modeling results and comparative analysis of the use of various algorithms of
swarm intelligence.
Keywords: Swarm intelligence · Drug design · Optimal immune network mod-
el.
1 Introduction
Nowadays modern methods of artificial intelligence are widely used in pharmacology
for computer modeling of new drug compounds with pre-defined properties. The
creation and investigation of chemical compounds is associated with the processing of
multidimensional data. The development of information technologies based on intel-
lectual approaches for processing and analysis a large data sets and solution of fore-
casting is an actual problem. Neural networks [1], genetic algorithms [2], artificial
immune systems (AIS) [3, 4], swarm intelligence algorithms [5] and others are widely
used in medicine. The article [6] presents the joint use of modified AIS and a partial
Copyright © by the paper’s authors. Copying permitted for private and academic purposes.
In: S. Belim et al. (eds.): OPTA-SCL 2018, Omsk, Russia, published at http://ceur-ws.org
350 G.A. Samigulina, Zh.A. Massimkanova
least square regression method for breast cancer diagnosis. The proposed approach
has a significant effect on the classification accuracy for clinical diagnosis and can be
used to solve detection problems. The research [7] describes artificial immune recog-
nition algorithm, which shows the highest classification accuracy for tuberculosis
diagnosis. The method can be applied for any diagnostics, the classification accuracy
will be high, especially for large data sets.
Forecasting quantitative structure-activity relationship (QSAR) of drugs and identi-
fying the links between compounds structure and their activity is an actual problem in
pharmacology. One of the important steps in the process of forecasting structure-
property/activity relationship is the selection of informative descriptors for reducing
the size of a descriptor space.
Nowadays bioinspired intellectual approaches to solve optimization problems are
actively developed. Swarm intelligence algorithms, such as ant colony optimization
(ACO) and particle swarm optimization (PSO), which based on animal and insect life
behavior to find the shortest path between food source and their nests, are applied in
QSAR modeling [8]. The article [9] describes optimization algorithms (evolutionary
algorithms, particle swarm algorithms) to determine the best position of a ligand in
protein-ligand docking. The research is executed using program AutoDock. The re-
sults show the effectiveness of particle swarm algorithm, especially for highly flexible
ligands. The work [10] presents a new approach, which is called two-step swarm in-
telligence. The main idea of the algorithm is to divide the heuristic search into two
stages. At the first stage, agents create solutions that are used as the initial states in the
second stage. This algorithm is used in joint with ACO and PSO algorithms. The
obtained experimental results demonstrate that the two-step swarm intelligence im-
proves the characteristics of used algorithms. In research [11], there is proposed PSO
algorithm to search an optimal number of features and reduce the dimensionality of
spectral image data. The study [12] deals with particle swarm and firefly algorithm
for diagnosing a tumor in images of magnetic resonance and computer tomography.
In article [13], there is studied a joint use of chaotic optimization algorithm and PSO
algorithm to improve the classification accuracy, which are used in the selection of
data sets with certain pharmacodynamic properties of drug. The experimental results
present that the proposed method has good learning performance, strong generaliza-
tion ability and classification accuracy.
At construction the optimal immune network model for forecasting QSAR of
drugs, there is relevant to use an ontological approach, which allows structuring the
input and output data, take into account the features of functioning and interconnec-
tion, and save the time and computing resources while developing the information
system. In paper [14], an improved PSO algorithm based personalized ontology mod-
el is described. The model creates personalized user profiles and finds information
about users from local repositories. The experimental results show that the proposed
PSO algorithm based personalized ontology model is effective in comparison with
other models. In article [15], there is introduced PSO algorithm based on semantic
relations and tested on the engineering applications. The experimental results demon-
strate that for small data sets the optimization ability of PSO algorithm based on the
semantic relations is better than classical algorithms. The algorithm allows finding the
351 Optimal Immune Network Model for Computer-aided Design of New Drugs
optimal value for short period. The work [16] provides PSO algorithm in ontology
repository for semantic web service selection.
The following structure of the article is proposed: Section 2 describes problem
statement of the research and the immune network technology for forecasting of
QSAR of chemical compounds. Section 3 presents the creation of ontological models
of swarm intelligence algorithms. Section 4 is devoted to the development an infor-
mation system of forecasting for conducting scientific researches “SIIM” (Swarm
intelligence for immune network modeling) and the description of the database of
sulfonamides with different duration of action. Section 5 presents the modeling results
using the chemical compounds of sulfanilamide group as an example and comparative
analysis of used algorithms. At the end of the article, conclusion and references are
presented.
2 Problem Formulation and Solution Methods
The problem statement is formulated as follows: it is necessary to develop an infor-
mation system of forecasting for conducting scientific researches ”SIIM” (Swarm
intelligence for immune network modeling) for creation an optimal immune network
model of drug compounds of sulfanilamide group based on swarm intelligence algo-
rithms: modified algorithms of ant colonies and particle swarms.
Definition: optimal immune network is a network constructed based on the weight
coefficients of the selected informative descriptors and most fully characterizing the
considered chemical compound. The criterion of optimization is the storage of
maximum information at a minimum number of descriptors [3, 17].
The intellectual technology for forecasting the properties of new chemical
compounds consists of the stages of preliminary data processing, immune network
training, image recognition, energy error estimation and selection of candidates of
drug compounds [18]. Preliminary data processing includes the description of
chemical compounds in the form of descriptors, normalization, verification of the
completeness and reliability of descripts, also the reduction of low informative
descriptors. The selection of informative descriptors is performed using swarm
intelligence algorithms based on multi-algorithm approach [19], which allows to use
several algorithms.
The development of integrated ontological model (OM) allows to study of subject
domain of AIS in detail and to analyze of swarm intelligence algorithms deeply. The
use of modern ontological editors and the creation of OM facilitate the solution of
problem of selection informative descriptors and the construction of an optimal im-
mune network model. As a tool for developing OM, there is chosen the ontology edi-
tor Protégé [20].
352 G.A. Samigulina, Zh.A. Massimkanova
3 The Creation of Ontological Models
The integrated OM of immune network technology, which consists of OM of prelimi-
nary data processing, OM of image recognition and OM of energy error estimation of
AIS has been proposed. The integrated OM is presented in the form of a tuple of sets:
ОМINT = ,
where OMPR – OM of preliminary data processing;
OMIR – OM of image recognition based on AIS;
OMEEE – OM of energy error estimation of AIS.
The ontological model of preliminary data processing consists of ontological
models of algorithms of ant colony and particle swarms:
ОМPR = ,
where OMACO – OM of algorithms of ant colony;
OMPSO – OM of algorithms of particle swarms.
There are many modifications of classical swarm intelligence algorithms for pre-
liminary data processing. Ant colony algorithm has several modifications such as
AntSrank, Max-min ant system, Elitist ant system, etc. In addition, particle swarm
algorithm has following modifications: CoPSO, Fully informed PSO, Inertia
Weighted PSO, etc. Table 1 shows OM of swarm intelligence algorithms, OM of
image recognition, OM of energy error estimation.
Table 1. Content of ontological models.
Ontological model Content
Ontological model of Algorithms of ant colony:
algorithms of ant – AntSrank algorithm.
colony – Max-min ant system algorithm.
– Elitist ant system algorithm.
– Ant-Q algorithm.
– Classical BasicACO algorithm:
Generation of population size of agents.
Random permutation of agents.
Initialize the amount of pheromone.
Calculation of fitness function.
Determination the amount of pheromone.
Permutation of agents.
Delay the amount of pheromone.
Update local and global amount of pheromone.
Check stop condition.
Save global best position (gbest).
353 Optimal Immune Network Model for Computer-aided Design of New Drugs
Table 1 (continued)
Ontological model of Algorithms of particle swarms:
algorithms of particle – CoPSO algorithm.
swarms – Fully informed PSO algorithm.
– Inertia Weighted PSO algorithm.
– Time-Varying Inertia Weighted PSO algorithm.
– Classical BasicPSO algorithm:
Generation of population size of agents.
Generation of random position and random velocity of agents.
Calculation of fitness function.
Determination the best position of agents.
Migration of agents.
Update position and velocity of agents.
Update the best position of agents.
Check stop condition.
Save global best position (gbest) of agents.
Ontological model of Implementation of AIS technology and the creation of an optimal
image recognition immune network model [21, 22]:
- creation of matrixes of standards and matrixes of images formed
from time series (descriptors).
- training of AIS with the teacher.
- singular value decomposition (SVD).
- determination of binding energies between formal peptides.
- solution of the problem of image recognition based on the
determination of the minimum value of the binding energy and
forecasting.
Ontological model of Energy error estimation of AIS [21, 23]:
energy error estima- - Averaging of the potentials by the homologies.
tion of AIS - Calculation of the average amount of standard deviations between
the native structure energy and the energy of the randomly chosen
stacking of the chain.
- Determination of prediction risk factors.
Swarm intelligence algorithms at selection of informative descriptors show different
results depend on the size and quality of data, the availability of independent parame-
ters and the optimality criteria. There are no universal algorithms for preliminary data
processing. The advantage of using a multi-algorithmic approach is the possibility of
choosing swarm intelligence algorithm, which allows to create an immune network
model with the best prognostic properties and shows the minimum value of generali-
zation error of AIS. Image recognition and energy error the estimation of AIS are
described in work [21].
354 G.A. Samigulina, Zh.A. Massimkanova
4 Information System of Forecasting for Conducting Scientific
Researches «SIIM»
Information system of forecasting for conducting scientific researches «SIIM» [24] is
used for selection informative descriptors at preliminary data processing. The infor-
mation system is developed in programming language Python 3.6. At the first step of
the information system, there is connected a database of descriptors of chemical
compounds. The database is displayed on the left screen of the interface. At next step
swarm intelligence algorithm (ant colony optimization or particle swarm
optimization) is chosen. The fields for input coefficients is displayed. A coefficients
are introduced depending on selected algorithm. For example, for particle swarm
optimization algorithm the coefficients are as followings: population size, iteration
numbers, weight and velocity. After introducing all coefficients it is need to click
"Run" button for calculation. The processing of multidimensional data is performed,
the allocation of informative descriptors and the construction of an optimal immune
network model are implemented. Modeling results are displayed on the right screen of
the interface. By comparing the results of AIS prediction, there is chosen swarm
intelligence algorithm with the minimum value of generalization error.
As the database has been used a database of descriptors of sulfanilamide group
with pre-defined pharmacological properties on the basis of the resource Mol-instincs
and PubChem [25]. The database consists more than 1500 descriptors. Table 2 shows
a fragment of the sulfanilamide database, which are classified into short acting,
medium acting and long acting sulfanilamides.
Table 2. A fragment of the sulfanilamide database.
Сlass Number Relative Relative Relative Relative … Relative
of number number number number number
atoms of C of H of O of N of aro-
atoms atoms atoms atoms matic
bonds
short_acting 27,00 0,37 0,37 0,07 0,15 … 0,43
medium_acting 24,00 0,33 0,42 0,13 0,08 … 0,13
short_acting 31,00 0,35 0,42 0,10 0,10 … 0,19
long_acting 35,00 0,34 0,40 0,11 0,11 … 0,31
long_acting 36,00 0,42 0,39 0,06 0,11 … 0,41
short_acting 33,00 0,36 0,42 0,06 0,12 … 0,29
medium_acting 28,00 0,36 0,39 0,11 0,11 … 0,32
short_acting 46,00 0,33 0,54 0,04 0,07 … 0,17
… … ... … … … … …
long_acting 30,00 0,37 0,40 0,07 0,13 … 0,29
355 Optimal Immune Network Model for Computer-aided Design of New Drugs
5 Modeling Results and Comparative Analysis
At modeling sulfanilamides based on PSO algorithm the population size is 100,
iteration number is 50, с1 (weight) = 1, с2 (velocity) = 2, report frequency equal to
50. As a result, there were selected 49 informative descriptors from 1500 ones. Figure
1 shows the graph of selected informative descriptors based on PSO algorithm.
Fig. 1. The visualization of the selected informative descriptors based on PSO algorithm with a
population size of 100
If at modeling the population size is 200, then, as a result, there are selected 11
informative descriptors from 1500 ones (Fig. 2).
Fig. 2. The visualization of the selected informative descriptors based on PSO algorithm with a
population size of 200
356 G.A. Samigulina, Zh.A. Massimkanova
From comparison purposes, both algorithms have the same iteration numbers and
population sizes. The comparative analysis allows to define optimization algorithm
with the best performance and low execution time. Table 3 shows a comparison of
modeling results based on algorithms of ant colonies and particle swarms.
Table 3. Comparative analysis of the sulfonamides modeling results
Coeffecients Ant colony algorithm Coeffecients Particles swarm
algorithm
Population size 100 200 Population size 100 200
Iteration numbers 50 Iteration numbers 50
Amount of 1 Weight 1
pheromone
Evaporation of 2 The velocity of 2
pheromone the particle
Report frequency 50 Report frequency 50
Sulfonamides modeling results
General amount 1500 1500 General amount 1500 1500
of descriptors of descriptors
The amount of 25 17 The amount of 49 11
informative informative
descriptors descriptors
The amount of population size and iterations affect the effectiveness of swarm intelli-
gence algorithms. A large number of populations and iterations allows agents to ex-
plore descriptor space more detailed and reduce the number of selected informative
descriptors. The running time of PSO algorithm is 8 seconds when the population size
is equal to 100 particles. The running time of ACO algorithm is 20 seconds with
population size of 100. The modeling results show, PSO algorithm is considered as best
optimization method with minimum execution time.
6 Conclusion
Therefore, the study of structure-property/activity relationship of drug compounds,
the development of new non-traditional intellectual approaches of QSAR and the
computer molecular design of drugs with pre-defined properties are one of the most
actual and main tasks of modern pharmacology aimed at reduction the time and cost
of creating new drugs. The developed information system for conducting scientific
researches “SIIM” with the use of immune network technology for forecasting struc-
ture-property/activity relationship of chemical compounds and multi-algorithmic ap-
proach, which allows integrating different methods of artificial intelligence to solve
the problem of computer molecular design of new drugs with pre-defined properties.
The application of multi-algorithmic approach with the use of modified algorithms of
ant colonies and particle swarms allows performing preliminary data processing effi-
ciently, allocating the informative set of descriptors and creating an optimal immune
357 Optimal Immune Network Model for Computer-aided Design of New Drugs
network model. At developing an information system, the use of ontological models
allows structuring data and analyzing the hidden interactions between descriptors. The
advantage of immune network modeling technology is the energy error estimation by
homologues [21], which allows to separate chemical compounds with almost identical
structure, but belonging to different classes of pharmacological activity.
The work was carried out according to the grant of the CS MES RK on the theme:
"Development and analysis of databases for the information system for prediction the
"structure-property" dependence of drug compounds based on artificial intelligence
algorithms" (2018-2020).
References
1. Zamani, M., Kremer, S.C.: Neural networks in bioinformatics. In: Handbook on Neural In-
formation Processing. vol. 49, pp. 505-525, Springer-Verlag, Berlin Heidelberg (2013)
Wong, K.: Evolutionary Algorithms: Concepts, Designs, and Applications in Bioinformat-
ics. Nature-Inspired Computing: Concepts, Methodologies, Tools, and Applications,
https://www.igi-global.com/chapter/evolutionary-algorithms/161025, [accessed 23 March
2018]
2. Tarakanov, A., Nicosia, G.: Foundations of immunocomputing. In: Proceedings of the 1-st
IEEE Symposium on Foundations of Computational Intelligence (FOCI’07). pp. 503–508.
IEEE, Honolulu (2007)
3. Dasgupta, D.: Recent advances in artificial immune systems: models and applications.
Applied Soft Computing Journal 11, 1574–1587 (2011)
4. Das, S., Abraham, A., Konar, A.: Swarm intelligence algorithms in bionformatics. Studies
in Computational Intelligence (SCI) 94, 113-147 (2008)
5. Zhao, W., Davis C.E.: A modified artificial immune system based pattern recognition ap-
proach an application to clinic diagnostics. Artificial Intelligence in Medicine 52(1), 1–9
(2011)
6. Shamshirband, S., Hessam, S., Javidnia, H., Amiribesheli, M., Vahdat, S., Petković, D.,
Gani, A., Kiah, M.L.: Tuberculosis disease diagnosis using artificial immune recognition
system. International Journal Medical Science 11(5), 508–514 (2014)
7. Goodarzi, M., Dejaegher, B.: Feature selection methods in QSAR studies. Journal AOAC
Int. 95(3), 636–651 (2012)
8. Guo, L., Yan, Z., Zheng, X., Hu, L., Yang, Y., Wang, J.: A comparison of various optimi-
zation algorithms of protein-ligand docking programs by fitness accuracy. Journal Molecu-
lar modeling 20(7). http://doi.org/10.1007/s00894-014-2251-3, [accessed 20-January-
2018]
9. Gómez, Y., Bello, R., Puris, A., García, M.M.: Two step swarm intelligence to solve the
feature selection problem. Journal of Universal Computer Science 14, 2582–2596 (2008)
10. Monteiro, S.T., Kosugi, Y.: Applying particle swarm intelligence for feature selection of
spectral imagery. http://citeseerx.ist.psu.edu/doi=10.1.1.726.8470, [accessed 20 November
2017]
11. Thamaraichelvi, B., Yamuna, G.: Hybrid firefly swarm intelligence based feature selection
for medical data classification and segmentation in SVD - NSCT domain. International
Journal of Advanced Research. http://dx.doi.org/10.21474/IJAR01/1544, [accessed 10
January 2018]
358 G.A. Samigulina, Zh.A. Massimkanova
12. Liu, F., Zhou, Z.: A new data classification method based on chaotic particle swarm opti-
mization and least square-support vector machine. Сhemometrics and intelligent laborato-
ry systems, 147–156 (2015)
13. Reha, S., Geetharani, M.: Swarm intelligence based fuzzy with personalized ontology
model for web information gathering. International Journal of Advances in Computer Sci-
ence and Technology 3(1). http://warse.org/pdfs/2014/ijacst07312014.pdf, [accessed 12
December 2017]
14. Shao, L., Bai, Y., Qiu, Y., Du, Zh.: Particle swarm optimization algorithm based on se-
mantic relations and its engineering applications. Systems Engineering Procedia 5, 222–
227 (2012)
15. Yaamini, S.K., Suganya, I.: Semantic Web service selection using particle swarm optimi-
zation. International Journal of Engineering Trends and Technology (IJETT) 4(603) (2013)
16. Samigulina, G.A.: Immune Network Modeling Technology for Complex Objects Intellec-
tual Control and Forecasting System: Monograph. Science Book Publishing House, USA
(2015)
17. Samigulina, G.A., Samigulina, Z.I.: Immune network technology on the basis of random
forest algorithm for computer-aided drug design. Bionformatics and biomedical engineer-
ing, 50-61, Spain (2017)
18. Samigulina, G.A., Samigulina, Z.I.: The construction of an optimal immune network mod-
el for forecasting the properties of unknown drug compounds on the basis of multi-
algorithmic approach. Problems of informatics 2, 21-29 (2013)
19. Samigulina, G.A., Massimkanova, Z.A.: Ontological models of swarm intelligence algo-
rithms for immune network modeling of drugs. Bulletin of Al-Farabi KazNU 1(93), 92-
104 (2017)
20. Samigulina, G.A.: Development of the decision support systems on the basis of the intel-
lectual technology of the artificial immune systems. Automatic and remote control. 74(2),
397-403, Springer (2012)
21. Samigulina, G.A., Samigulina, Z.I.: Development of the technology of immune network
modeling for computer-aided molecular design of drugs (computer program). Certificate of
state registration of the object of intellectual property No. 473, Astana (2011)
22. Samigulina, G.A., Samigulina, Z.I.: Energy errors estimation of artificial immune system
on homologues (computer program). Certificate of state registration of the object of intel-
lectual property No. 396, Astana (2008)
23. Samigulina, G.A., Massimkanova, Zh.A.: Swarm intelligence for immune network model-
ing “SIIM” (computer program). Certificate of state registration of the object of intellectu-
al property No. 319, Astana (2017)
24. https://www.molinstincts.com, [accessed 21 April 2017]