Bbs 061

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

B RIEFINGS IN BIOINF ORMATICS . VOL 14. NO 4. 411^ 422 doi:10.

1093/bib/bbs061
Advance Access published on 14 October 2012

Mathematical modeling of biological


systems
Santo Motta and Francesco Pappalardo
Submitted: 30th April 2012; Received (in revised form) : 23rd July 2012

Abstract
Mathematical and computational models are increasingly used to help interpret biomedical data produced by
high-throughput genomics and proteomics projects. The application of advanced computer models enabling the

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


simulation of complex biological processes generates hypotheses and suggests experiments. Appropriately inter-
faced with biomedical databases, models are necessary for rapid access to, and sharing of knowledge through data
mining and knowledge discovery approaches.
Keywords: mathematical biology; computational models; systems biology

INTRODUCTION data stored on biomedical databases through text


Revolutions in biotechnology and information tech- mining and knowledge discovery approaches.
nology have produced enormous amounts of data and Modeling is the human activity consisting of
are accelerating the process of knowledge discovery of representing, manipulating and communicating
biological systems. These advances are changing the real-world daily life objects. As one can easily realize,
way biomedical research, development and applica- there are many ways to observe an object or, equiva-
tions are conducted. Clinical data complements bio- lently, there are many different observers for the
logical data, enabling detailed descriptions of both same object. Any observer has ‘different views’ of
healthy and diseased states, as well as disease progres- the same object, i.e. ‘there is no omniscient observer
sion and response to therapies. The availability of data with special access to the truth’. Each different
representing various biological states, processes and observer collects data and generates hypothesis that
their time dependencies enables the study of biolo- are consistent with the data. This logical process is
gical systems at various levels of organization, from called ‘abduction’. Abduction is not infallible,
molecules to organism and even up to the population though; with respect to a scientific unknown, we
level [3–5]. Multiple sources of data support a rapidly are all blind.
growing body of biomedical knowledge, however, A system is a collection of interrelated objects. For
our ability to analyze and interpret this data lags far example, a biological system could be a collection of
behind data generation and storage capacity. different cellular compartments (e.g. cell types) spe-
Mathematical and computational models are increas- cialized for a specific biological function (e.g. white
ingly used to help interpret biomedical data produced and red blood cells have very different commitments).
by high-throughput genomics and proteomics pro- An object is some elemental unit upon which obser-
jects. The application of advanced computer models vation can be made but whose internal structure is
enabling the simulation of complex biological pro- either unknown or does not exist. The choice of
cesses generates hypotheses and suggests experiments. the elemental unit defines the representation scale of
Computational models are set to exploit the wealth of the system. A model is a description of a system in

Corresponding author. Santo Motta, Department of Mathematics and Computer Science, University of Catania, V.le A. Doria, 6,
95125 Catania, Italy. Tel.: þ39 095 7383073. Fax: þ39 095 330094. Email.: motta@dmi.unict.it
Santo Motta, Laurea in Physics (University of Catania,1970) and MSc in Applied Mathematics (University of London, Queen Mary
College, 1971). Associate professor of Mathematical Physics (Faculty of Pharmacy, Department of Mathematics and Computer Science,
University of Catania). His present scientific interests are BioMaths, BioComputing and BioInformatics.
Francesco Pappalardo is a researcher at the University of Catania. He was a visiting research scientist at the Dana-Farber Cancer
Institute in Boston and at the Molecular Immunogenetics Labs, IMGT in Montpellier. His major research area is computational
biology.

ß The Author 2012. Published by Oxford University Press. For Permissions, please email: journals.permissions@oup.com
412 Motta and Pappalardo

terms of constitutive objects and the relationships This article is organized as follows: in the next two
among them, where the description itself is, in gen- sections (Models of Systems and The Modeling
eral, decodable or interpretable by humans. Process), we describe the types of models and the
Generally speaking, a system is an unknown ‘black modeling processes in scientific investigations in a
box’ (S) which, under a specific external stimulus general context; then in the next section (Models
(input E) produces a response (output R) [19]. in Biology: Scales and Complexity) we go more spe-
Using this general definition, one can identify three cific and talk about models in biology and medicine;
primary scientific uses of models [12]: (i) synthesis or few examples of models are briefly shown in the
knowledge discovery; to use the knowledge of section ‘Tools and Applications’; finally we draw
inputs E and outputs R to infer system characteris- our ‘Conclusions’ in the last section.
tics; (ii) analysis and prediction; to use the knowledge
of the parts and their stimuli (i.e. the inputs E) to
MODELS OF SYSTEMS

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


account for the observed response (i.e. the output R)
and eventually, to predict response to different Not all scientific models are expressed in a precise,
stimuli. (iii) Instrumentation or device; to design an numerical and quantitative way. Actually, one can
‘alternative system’ (i.e. hardware or software), able identify four different types of models: verbal
to reproduce the input–output relationship with the models, conceptual or diagrammatic models, physical
best possible adherence to the studied system. models and formal models.
Secondary uses of models account for conceptual In this article, we focus mainly on diagrammatic
frameworks to design new experiments, methods to and formal models and we concentrate on the model
summarize or synthesize large quantities of data, tools building process.
to discover relationships among objects.
In this article, we analyze models and modeling Verbal models
processes specific for the biology. We mainly focus In verbal models the system is described in words.
on the use of models aiming at the points (i) and These models, based on observations, usually evi-
(ii) as tools for knowledge discovering in biology. dence in a simple way the objects and relations
The mathematical methods used in modeling bio- among the objects in the system. A verbal model is
logical systems vary according to different steps of the a rough and sometime ambiguous qualitative repre-
process. We focus on the mathematical representa- sentation of the knowledge of the system. These
tion of the system. However, other important steps kinds of models are used in the first approach to
in the modeling processes are parameters fitting and the analysis of biological system.
model selection. We will not analyze the mathem-
atical methods in those two important aspects as Conceptual or diagrammatic models.
these would require separate review papers. In conceptual or diagrammatic models the system is
Methods for parameters fitting refer to wide area of described by a graphical representation of the objects
mathematical optimization, whereas methods for and the relationships describing the underlying dy-
model selection mainly use statistical techniques. namical processes. To develop these kind of models
On top of these, sensitivity analysis and phase– the understanding of the available data needs to be
space analysis of the models may be required. sufficient to have a detailed (even if not exhaustive)
Interested readers may find more information in idea of the objects (or entities) and relations. A con-
these references: [11, 15, 20]. ceptual model (CM) represents ‘concepts’ (objects or
Models for technical use are formal models, but entities) and relationships between them. In com-
the strategy for building them is quite different and puter science, CM are also referred to as domain
therefore, we leave them out of the present discus- models. A CM is expressly independent from the
sion. In the following we will refer to this type of design and free from implementation concerns.
models as Black Box Models (BBM). It is worth The aim of a CM is to convey the meaning of
pointing out that, as we will mention later on, alter- terms and concepts used by ‘domain experts’ to ra-
native systems can be considered parts of a large tionalize the problem and to find relationships
model to account for effects whose origin can be among the different concepts. The CM aims to clar-
neglected without compromising the understanding ify the meaning of the usually ambiguous terms to
of the whole phenomena. minimize as much as possible problems arising from
Mathematical modeling of biological systems 413

different interpretations of terms and concepts. If a different parts or compartments (i.e. organs) with
‘domain ontology’ is available, then the meaning of different functions. In this case, ‘Compartment
the variety of terms used should be linked to it. Once models’ are widely used and each compartment
the domain concepts have been modeled, the model may pick a different mathematical representation.
becomes a stable basis for subsequent development of Models can also represent physical variables in
applications in that specific domain. Furthermore, different ways. Besides, the model may or not con-
the concepts behind the conceptual model can be sider the evolution of the system with respect to time
mapped to physical design or implementation con- (dynamic versus static models). Time can be treated
structs using either manual or automated code gen- as continuous or discrete variable (time-continuous
eration approaches. versus time-discrete models). Likewise, spatial distri-
A conceptual model can be described using vari- bution of objects in each compartment may or
ous notations, such as Unified Modeling Language not be relevant (spatially-heterogeneous versus
(UML) [34, 9], Object Modeling Technique (OMT)

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


homogeneous models). Finally, similar objects
for object modeling [32], or Information Engineer- may be treated as individuals or taken in bulk
ing or Integration Definition for Information Mod- (particle models versus population models). In the
eling (IDEF1X) for Entity Relationship Modeling first case, individual objects are identified by a
[6]. In UML notation, the conceptual model is unique state or by a large, but finite number of
often described with a class diagram in which classes states (one-state particle versus finite-state automata).
represent concepts, associations represent relation- Lastly, the relations between objects can be described
ships between concepts and role-type of an associ- as deterministic or stochastic rules (deterministic
ation represents role types taken by instances of the versus stochastic models).
modeled concepts in various situations. According to the different modeling choices, one
can get single versus multicompartments models,
Physical models including transport, evolutionary differential equa-
In physical models the representation is done using a tions versus algebraic equations or spatial partial dif-
mock-up of a real system or object (like a scale ferential equation, differential equations versus
model of an aircraft or of a ship). These type of difference equations, ordinary differential equations
models are mostly of interest for engineers. They (ODE) versus partial differential equations (PDE),
are widely used when the properties of the system kinetic methods, agent-based methods (ABM) or
are almost ‘scale-invariant’, i.e. independent from the cellular automata (CA) versus ODE or PDE; deter-
size of the physical model built to represent the real ministic methods (ODE or PDE, etc.) versus stochas-
system to produce smaller-scale prototypes. tic methods (stochastic ODE and PDE).

Formal models Statistical and artificial intelligence-based models


Formal models represent the knowledge of the A statistical model is a formalization of the relation-
system using mathematical structures. The mathem- ships between variables (i.e. object’s measurable
atical representation of the model depends on the characteristics) in the form of mathematical equa-
knowledge of the system, on some modeling choices tions, the only difference with the mathematical
(for instance, the spatial scale of representation) and models described above is that in statistics, all vari-
the aim of the modeling process. There are a large ables and/or parameters of the model include a level
variety of mathematical/computational methods that of uncertainty. When the relationship between two
can be used and the selection of the proper one fol- objects is too complex to be easily guessed, one can
lows rules that are often matters of experience. At a resort to probabilistic measures and statistical or arti-
first glance, there are few questions one may ask ficial intelligence methods to reproduce the response
to address the choice of the proper mathematical/ relationship (see e.g. refs. [21, 22, 37]). In these kind
computational method. Those questions are mainly of models, the detailed analysis of the system com-
related to the description of the system with respect ponents is usually ignored because, the objective of
to its different parts or components, the physical vari- the model is limited to reproduce the system stimu-
ables space and time, the type of relations between lus/response relation. Examples of this approach are
objects and the object representation perse. In systems the lumped models using equivalent circuits, neural
biology, a system is viewed as an assembly of network, etc.
414 Motta and Pappalardo

THE MODELING PROCESS system we wish to model, as some informations may


The modeling process consists of the following steps: be more relevant than the others.
(i) model implementation consisting in describing by
a formal language the objects/relationships identified Current knowledge
in the system under study using a mathematical A second crucial step in the modeling process is to
structure and/or a computer code; (ii) use the collect the knowledge on the system under investi-
model to forecast the system behavior and (iii) evalu- gation. This is conducted by consulting the scientific
ate the model adherence to reality by matching pre- literature, including experimental reports and/or
dictions with available data. discussing with field-experts. In the biomedical
To find a good model is an issue. Modeling is a field, data sets of literature record (e.g. Pubmed)
hard problem in itself and failure is not a rare event. can highly facilitate the task of browsing the vast
The modeling procedure is a process in itself that amount of information available nowadays. In this

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


follows a semi-formal set of rules. The methodology respect, methods of data mining and data extraction
lean on four macro steps [31]: (i) understand the may be very useful [17].
problem, i.e. clearly define the questions one asks
to the model; (ii) devise a plan for solving the prob-
lem, i.e. define a series of steps to be put in practice Model structure
to find an adequate model of the system under in- A model is a representation of a real system and has
vestigations. This step includes knowledge and data its own structure. For the sake of manageability the
acquisition from field experts and literature, model model structure should include, of all the acquired
structure, model hypothesis, conceptual model, knowledge, only those considered to be relevant for
choosing the appropriate mathematical formalism, the purpose of the study (‘Realism’); the level of
solving the formal model, get the results, check details of the model results should also be determined
model results matching to available data and so on; a priori (‘Precision’); finally, a model can be general,
(iii) execute the plan, i.e. perform the steps in (ii) and that is applicable to other similar systems or specific
(iv) check the correctness of the answer and eventu- to the system of interest (‘Generality’). Realism,
ally refine the model. This last point is a major test to Precision and Generality are competing properties.
evaluate the hypothesis formulated when setting the Each of these properties trades-off against the other
model. two. Deciding a model structure is to find a proper
As mentioned before, we are mostly interested in balance between those competing properties, which
models for the analysis and predictions. For these satisfy the model objectives. Decision on the model
models, the classical description of modeling process structure is crucial for defining the model hypothesis,
is shown in Figure 1. It is worth to mention that the diagrammatic model construction and mathematical
schema illustrated in Figure 1 does not have the pre- formulation.
tense to be the most general one: it is a general ap-
proach that can be used in the analysis and Hypotheses
predictions models. The next step in the modeling process is to translate
objectives and current knowledge we wish to
Model objectives include in the model in a list of specific working
As we already pointed out, a proper definition of the hypothesis. These are usually verbal statements, but
model objectives is a fundamental step as it implies a could also be quantitative relationships. Working
certain level of comprehension of the problem. The hypotheses are the basis of the model we are going
reason for building a model should by clear and a to built and model results will depend on them. In
proper clarification of the objective must answer to doing the cycle refinement of model, the starting
major questions: (i) what is the system to be mod- hypothesis should be critically, repeatedly analyzed.
eled; (ii) what are the major questions to be ad-
dressed by the model; (iii) How good must the Conceptual model
model be and to what it will be compared with? The conceptual model is a graphical representation
(iv) How we will analyze and use the model output? of the relevant system knowledge and model object-
All these questions need to be clarified before we ives that have been identified in the hypotheses. In
proceed in searching the current knowledge on the the conceptual model compartments, objects and
Mathematical modeling of biological systems 415

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


Figure 1: The description of the modeling process. The top part of the figure refers to the formulation of the
model, i.e. identification of the model objectives; collection of the current knowledge about the biological system
under investigation; choosing the most appropriate model structure to satisfy the model objectives; translate
objectives and knowledge into model hypotheses; draw a conceptual model; identify mathematical technique and
develop the formal model. Central part of the figure refers to calibration i.e. estimating and fitting model param-
eters. Finally, the bottom of the figure refers to analysis and evaluation, i.e. comparison of the model results against
experimental data sets and analysis of discrepancies.

relations will be described in a diagram where the set loose relations in strict mathematical terms. Notice-
of objects are fully clarified and relations bounded. ably, model objectives play an important role
because, a detailed description of the biological
Mathematical formulation system may turn out to be useless if not required
This is usually the trickiest part of the modeling pro- by the model objectives.
cess, requiring the choice of a mathematical struc- Choosing a mathematical formulations is a map-
ture, which is appropriate for the model objectives ping of the model into the mathematical domain to
and is able to describe in quantitative form the obtain a formal model. A good formal model must
hypotheses. This step of the process requires a certain be a compromise between the competing properties
level of mathematical sophistication and, more im- of any model (Realism, Precision and Generality)
portantly, it requires to define vague concepts and and should take into account some specificity of
416 Motta and Pappalardo

the mathematical domain. Accordingly, we can need to be carefully chosen. Models of biological
identify the major properties of formal models as systems can involve scores of dynamic variables and
follows: ‘relevance’, capturing the essential properties thousands of parameters, especially when spatial pro-
of the phenomenon; ‘computability’, transfer model cesses are investigated. In this respect, checking the
hypothesis into a mathematical/computational infra- computer results to match available data is not a triv-
structure that can be solved to give the desired results ial exercise.
with the required precision; ‘understandability’,
offering a conceptual framework for thinking about
the scientific domain; ‘extendibility’, allowing the Parameter estimation and tuning
inclusion of additional real-world objects in the Once the model formulation has been translated on a
same mathematical scheme. Taking into account computer, a further step is necessary before a simu-
the biological complexity, a very detailed model of lation can be run. Complex models usually contain
many parameters whose numerical value must be

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


a biological system may work out to be unsuitable to
automatic resolution, i.e. not computable (for in- determined. In biology this may not be an easy
stance, too many equations to be solved or too task as numerical values of parameters, if directly
many unknown parameters to be estimated); in con- measurable, are often known with a large uncer-
trast, a too simplistic model may not be able to ac- tainty. Methods for parameter fitting are mainly
count for the complexity of the biological system of based on optimization techniques that minimize
interest. the difference between real experimental data and
A good formal model must be understandable by model output. There are a variety of techniques ran-
domain experts so that they could use it for their ging from stochastic methods and gradient descent
own quantitative reasoning. Finally, as biology is a methods [11]. In some cases, specific experiments
fast growing science, extensibility is an important may be required for estimating parameters (see e.g.
characteristic for biological models. When new ob- the model of influenza described in ref. [30]).
jects and relations in the system are derived from However, a model can contain parameters that
laboratory experiments, it should be easy to extend cannot be estimated a priori. In these cases one can
the model with minor changes in the mathematical have an a posteriori estimation. This means that the
structure. values of the parameters can be estimated by another
In most cases models are equipped with just one model made ad hoc just to fit the data available [14,
mathematical structure of those mentioned previ- 29].
ously. However, there are also ‘hybrid’ models
where different mathematical structures are used in Model validation and cyclic refinement
combination but also models which add to a detailed Comparison of model results, or simulations are the
description of the system of interest other types of final part of the modeling process. The general goal
models (e.g. BBM), which mimic the effects of other of a model is to reproduce data from observation or
systems interlinked with the one under from experiments (descriptive models) or to predict
investigations. the result of new observations or experiments (pre-
We will describe some hybrid models in the sec- dictive models). Obviously, results need to be vali-
tion ‘Tools and Applications’. dated according to model’s objectives. In some cases,
a qualitative agreement between model results and
From mathematical formulation to the experimental data is adequate, in other cases, a quan-
numerical solution titative agreement is necessary. It is a common prac-
Only very simple models can be analyzed analytically tice for model validation to require the model results
(i.e. by algebraic derivation of the system properties). to be validated against independent data sets. Model
In most cases the model is either directly imple- results that fail to fit the experimental data set not-
mented as a computer code (i.e. the algorithm-like withstanding changes in the model-free parameters,
in ABM) or equations must be solved numerically. suggest a further model refinement. In this respect, in
Even if there are a variety of methods for solving the way back and forth between model refinement
equations, transferring equations into computer and data validation one can discover interesting
code is a possible source of error and appropriate properties of the system of interest. The process
method to avoid errors due to numerical instabilities itself leads to discovering of new knowledge.
Mathematical modeling of biological systems 417

MODELS IN BIOLOGY: SCALES AND processes. Cellular scale is thus highly connected
COMPLEXITY with the sub-cellular scale but, modeling at this
As already pointed out, any natural phenomena can scale, one may forget the details of single cell
be observed at different scales thus, in describing the models and consider them as BBM. The areas of
phenomenon by conceptual and quantitative models mathematical methods and tools involved in this
one needs to choose to appropriate scale to describe description refer to statistical mechanics, cellular
the experimental data available. automata, lattice gas and other similar approaches.
However, in almost all complex natural phenom- The ‘macroscopic scale’ include tissues, organs
ena there are aspects that cannot be even observed at (i.e. a collection of tissues joined in structural unit
just one scale of description (either temporal or spa- to serve a common function), systems (i.e. a group of
tial). To study these specific facets of reality, multi- organs working together to perform a certain task)
scale models that represent objects and relationships and organism.

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


on different levels of abstraction are required. In this scale, one is interested in describing the
Choosing a scale depends on which aspects of the dynamical behavior of observable quantities, in
phenomena, from ‘micro’ to ‘macro’, one is inter- most cases, the concentrations of various entities
ested to analyze. In physics this is already a well (cells or molecules). Tissues are usually described
defined approach that originates from different re- using techniques originating from physical continu-
search areas, and the distinction between different ous systems, i.e. ordinary or partial differential equa-
scales is based upon the characteristic lengths of ob- tions or moments of kinetic equations. In describing
jects and the characteristic time of the phenomena organs, a model is required to describe both the main
under investigation. For instance, microphysics refers tissue and the sporadic tissues but also, and most im-
to areas of physics that study phenomena, which take portantly, the biological function. To model a
place on the microscopic scale (length scales <1 mm), system, one is required to consider a network of
such as: molecular physics, atomic physics, nuclear organs that perform a specific task. Depending on
physics and particle physics. the modeling goal the model of a biological system
In the life sciences the definition of a scale is a bit can be arranged with different levels of details.
more ambiguous. A basic unit available for defining a Organs can be fully described in their components
scale is the ‘cell’ with no regard to its physical di- or simply as BBM performing a given task.
mension. Starting from this, one can define different Connections between organs (like lymphatic vessels)
scales: the ‘sub-cellular’ or ‘intra-cellular’ scale, the can be described physically (dynamical description of
‘cellular’, mesoscopic or ‘inter-cellular scale the the fluid motion in the vessels) or simply considering
‘macroscopic scale’ and the ‘populations scale’. the flux and the time required to move portions of
Models developed at sub-cellular scale deal with fluid from different organs, i.e. though law of
the evolution of the physical and biochemical state of transport.
a single cell. This scale involves genes, proteins and Finally, in the population scale, one is interested in
signals in cell nucleus and surface, which regulate the describing the dynamics of the populations with re-
evolution of the cell and any signaling processing spect to one or more characteristics. Models of
operations of the cells enabling cell crosstalk. epidemics or population controls are well known
Modeling the overall activity of a single cell is a models at this scale. Population dynamics is ex-
very hard problem as many biological details of this tremely complex because the effects of all previously
activity are unknown. mentioned scales and the effects of the environment
Biologists and modelers have joined forces to de- on a single organism can modify the overall dynamic
velop and use mathematical and computer science of the populations. In this class of models, a single
techniques in modeling sub-cellular phenomena. organism can or cannot be described in detail, ac-
Interested readers can find plenty of references in cording to the size of the populations one is required
the scientific literature [13]. to describe. In both cases, changes of the major char-
In the cellular scale, one is interested in describing acteristics of a single organism must be taken into
the evolution of a system consisting of a large account. For example, to describe the response of a
number of different interacting cells and molecules. population to large-scale vaccinations (as required in
Cell interactions are regulated by signals emitted and influenza epidemics), one does not describe in detail
received by cells through complex recognition any single organism but it may be required to
418 Motta and Pappalardo

consider the age structure of the population and the not an easy task. Phenomena occurring at different
effects of environments. At variance for small size scales have usually different characteristic time scales
populations, like modeling the effect of a new vac- and models’ output should be properly fitted.
cine for a small trial, a sufficiently detailed description Interested readers are referred to another study [3]
of the organism and the effect of the vaccine on each in this issue.
organism should be required. A variety of different
techniques are available for these classes of models. In
the former case, one uses both, ordinary or stochastic TOOLS AND APPLICATIONS
differential equations to describe the populations Whether we investigate the growth and interactions
dynamics or agent-based models (simple agents of an entire population, the evolution of DNA
representing a single organism) to study the result- sequences, the inheritance of traits, the spread of dis-
ing complex phenomena. In the latter case, a ease or the immune system response to a pathogen,

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


more detailed description of the organism is biological systems are marked by change and adap-
required and the population dynamics may eventu- tation. Even when they appear to be constant and
ally be extracted from the dynamic response of each stable, it is often the result of a balance of tendencies
organism. pushing the systems in different directions. The
choice of the mathematical approach depends on
Complexity and multiscale models the biological system one would like to model. In
Living organisms are complex systems. Using a ‘clas- this section, we sketch several applications of math-
sical’ definition, a complex system is a system com- ematical techniques that have been effective in
posed of different interconnected parts that, as a reproducing and in providing new insights of a par-
whole, exhibit one or more properties which do ticular biological problem.
not obviously arise from the properties of the Due to their incredible complexity, models that
individual parts. System complexity may be either a deal with an entire biological system are, to date,
‘disorganized complexity’ or an ‘organized complex- very few and actually incomplete. Instead there are
ity’. In the former case complexity arises from a very several mathematical models that act toward single or
large number of parts, whereas in the latter case, group of components of a biological system.
complexity is intrinsic to the system, eventually
with a limited number of parts, and its connections
rules. Tools for bioinformatics and systems
In living organisms both situations occurs. A living biology
organism is formed by a collection of different parts Signaling network has a key role in cellular physi-
which are, each of them, organized complex systems. ology and therefore, it has been widely studied in
Cells, organs, systems of the human body are each of several organisms. This is due mainly because all cells
the complex systems. interact and respond to the environment in which
As an example, the immune system is one of the they live. The bad news is that such networks are
very complex one where complexity arises both very complex due to their combinatorial explosion
from a very large number of parts (organs), constitu- nature. For this reason, frameworks for mapping
ents (cells and molecules) and rules hierarchically signal-transduction networks that avoids the com-
connecting different scales of the parts. binatorial explosion in some way are particularly
Models including many scales of a phenomenon needed. In ref. [35], a framework for mapping, visu-
are now requested both for knowledge discovery and alization and automatic model creation of
drug discovery. In life sciences not only an entire signal-transduction networks is presented, along
living organism, but also parts of the organism are with an example of its use to compile the, presently,
too complex to be represented in a single, precise, most comprehensive map of the yeast MAP kinase
multiscale model. The resulting model would cer- network.
tainly not be computable. Thus, one is forced to It is well known that the explosion of data origi-
break the conceptual model in a set of models nated from biology has made it increasingly import-
describing only part of the phenomenon (like a ant to provide metadata along side the core data
single organ, or a definite scale) and connect their itself. The concept of metadata derives from cart
outcomes [16]. To link models at different scales is a catalogs and libraries by describing the contents
Mathematical modeling of biological systems 419

and context of data files, the quality of the original Applications


data/files is greatly increased. This metadata may Immunology
comprise domain-specific information as described The role of mathematical modeling in immunology,
by minimal information checklists meant to enable one of the most complex fields in biology, were
accurate data reuse or may be ontological in nature, recognized early, beginning from the 1960s and
specifying more precisely the kind of entities under the 1970s. Since then, mathematical models have
consideration. The Minimum Information Required been used in various domains of immunology [23].
in the Annotation of Models Registry (http://www One of the major issues in vaccine and other im-
.ebi.ac.uk/miriam) provides unique, perennial and munologic approaches’ research is the testing of the
location-independent identifiers for data used in relevant biological variables when each experiment
the biomedical domain. In [18], the authors describe lasts 1 year. One clear example is the scheduling of
the new Identifiers.org service (http://identifiers.org) prolonged vaccinations. It is desirable to reduce as

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


that is built upon the information stored in the much as possible the number of vaccine administra-
Registry and which provides directly resolvable tions, e.g. to reduce the risk of side effects in humans.
identifiers, in the form of Uniform Resource In ref. [26] the authors describe the use of a math-
Locators (URLs). In the same context, the Systems ematical model based on ABM that faithfully repro-
Biology Graphical Notation (SBGN) facilitates duce in silico the behavior of a cancer-preventive
the representation and exchange of complex biolo- vaccination, suggesting a possible optimized vaccine
gical knowledge in a concise and unambiguous schedule [27] and highlighting certain critical issues.
manner [36]. In particular, although vaccinations could be reduced
Of specific importance are the major synthetic biol- in numbers without sacrificing efficacy, the intensity
ogy platforms, for managing the data, which is used to of early vaccinations was a key determinant of
create synthetic biological systems and to provide long-term tumor prevention needed for predictive
mechanisms to begin the process of creating standar- utility in the model. Moreover, long-term studies
dized data, algorithms and methodologies for syn- confirmed predictions of in silico modeling in which
thetic biology. Here we mention System Biology an immune plateau phase, once reached, could be
Workbench (http://sourceforge.net/projects/sbw/), maintained with a reduced number of vaccinations;
TinkerCell (http://sourceforge.net/ projects/tinker- revealing that the accuracy of mathematical model-
cell), Kappa and all the tools there provided (http:// ing of early immune responses is critical. This key
kappalanguage.org/), SBGN (http://www.sbgn. example shows that an integrated in vivo^in silico
org/), SBML (http://sbml.org/), Synthetic Biology approach could improve both mathematical and bio-
Open Language (http://www.sbolstandard.org/), logical models of cancer immunoprevention. An ex-
Clotho (http://clothocad.org/) and BEL ample of both qualitative analysis of the asymptotic
Framework $ http://belframework.org/) behavior and numerical simulations using nonlinear
Next generation sequence analysis has become an ODEs is given by the authors [2], where the math-
important task both in laboratory and clinical ematical modeling of the mammary carcinomaim-
settings. SeqAlto [25] is a new algorithm for mune system competition elicited by an external
read alignment. It is about to 10 faster than exist- stimulus is presented. A model for keloid formation
ing algorithms, while retaining high accuracy and triggered by virus, their malignant effects and
the ability to align reads with large (up to 50 bp) immune system competition have been described
indels. using a mathematical model developed by kinetic
Proteins execute and coordinate cellular functions theory of active particles described in a previous
by interacting with other biomolecules. Among these study [1].
interactions, protein–protein (including peptide- In ref. [28] the authors present a mathematical
mediated), protein–DNA and protein–RNA inter- model to analyze the co-stimulatory effects of
actions cover a wide range of critical processes and anti-CD137 monoclonal antibody (mAb) for the
cellular functions. Multi-VORFFIP [33] is a tool to melanoma treatment upon synergistic adoptive trans-
predict protein-, peptide-, DNA- and RNA-binding fer of activated OT-1 T cells. The reported in vivo
sites in proteins. One of its features is the web inter- experiments show that a single administration of
face to facilitate the use of the method and analysis of anti-CD137 mAb plus activated OT1 T cells is suf-
predictions to non-expert end-users. ficient to completely reject the B16-OVA, whereas
420 Motta and Pappalardo

single components or not activated OT1 T cells have and chemistry, i.e. genetics and biochemistry. Mole-
no success. The in silico experiments performed with cular biology provides the understanding of the
the presented computational model show very good interactions between the various systems of a cell,
agreement with their in vivo counterpart. As many including the interactions between the different
aspects of CD137 molecule biology are still not types of DNA, RNA and protein biosynthesis.
fully understood and the investigating of these as- Understanding how drugs and diseases are asso-
pects requires many difficult and expensive wet lab ciated in the molecular level, is of critical importance
experiments, the model is a good candidate for for better understanding of disease mechanisms and
becoming a predictive tool. treatments. Recently in a study [38], the authors
define a network-based gene closeness profile to
Circulatory system relate drug to disease and then develop a Bayesian
Circulatory system represents a biological system partition method to identify drug–gene–disease

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


made of organs that passes nutrients and other com- co-modules underlying the gene closeness data.
ponents to and from cells in the body to help fight Their mathematical approach and the related simu-
diseases, stabilize body temperature and pH, and to lations are applied to a set consisting of 723 drugs,
maintain homeostasis. Diseases associated with this 275 diseases and 1442 genes. It identified new drug–
system, i.e. cardiovascular diseases, have a major disease associations and highlighted their molecular
impact in Western countries. Mathematical models basis.
and numerical simulations of cardiovascular system Recently in another study [7], the authors deal
have been presented and have been demonstrated with drug resistance that has posed more severe
to provide help in understanding both their dy- and emergent threats to human health and infectious
namics and possible interventions. In a previous disease treatment. Due to less knowledge about the
study [10] the authors provide a general overview underlying mechanisms of drug resistance, wet-lab
of mathematical representation of vascular geome- only approaches achieved limited success. With the
tries extracted from medical images, the modeling use of interactome network of Mycobacterium tubercu-
blood rheology and the complex multilayer structure losis and gene expression data which are treated
of the vascular tissue, and its possible pathologies and with two kinds of antibiotics, the authors developed
the mechanical and chemical interaction between a mathematical workflow for giving new insights to
blood and vascular walls. bacterial drug resistance that can be gained by a sys-
tematic and global analysis of the bacterial regulation
Population dynamics network.
In another study [8], the authors describe and analyze Microbiology is the field of biology that studies
a periodically forced difference equation model for microscopic organisms i.e. bacteria, viruses, fungi,
malaria in mosquitoes that captures the effects of prions, protists and prokaryotes. There is a huge
seasonality and allows the mosquitoes to feed on a quantity of mathematical modeling contributions to
heterogeneous population of hosts. With the inte- this kind of biological systems, especially in the ana-
gration of the difference equation model with an lysis of the dynamics of pathogens. For example, in
individual-based stochastic simulation model for ref. [24], the authors use differential equations and
malaria in humans, they compare the effects of computational models to characterize the in vitro kin-
insecticide-treated nets (ITNs) and indoor residual etic behaviors of H5N1 avian, H1N1 seasonal and
spraying (IRS) in reducing malaria transmission, H1N1 2009 pandemic influenza virus strains. The
prevalence and incidence. They conclude showing approach provides relevant parameters for identifying
that ITNs are more effective than IRS in reducing and phenotyping potential pandemic strains.
transmission and prevalence, proving also that the
combination of both interventions is more effective
than either intervention alone. CONCLUSIONS
Biological systems are complex systems and the
Drug efficacy higher levels of complexity arise from collective
Molecular biology is the branch of biology that deals behavior and emerging properties at multiple
with the molecular basis of biological activity. This levels. This requires initially the analysis of large
field share knowledge with other areas of biology quantities low level data either acquired by direct
Mathematical modeling of biological systems 421

measurements or by accessing a variety of sources. 10. Formaggia L, Quarteroni A, Veneziani A. Cardiovascular


Mathematics: Modeling and Simulation of the Circulatory System.
These data then need to be integrated into various Milano, Italy: Springer, 2009.
network models or multiscale models. Models are a 11. Ghosh S, Matsuoka Y, Asai Y, et al. Software for systems
fundamental step in the scientific discovery. In this biology: from tools to integrated platforms. Nat Rev Genet
article, we described different types of models that 2011;12:821–32.
have been used in biology for knowledge discovery 12. Haefner JW. Modeling Biological Systems: Principles and
Applications. New York, USA: Springer, 2005.
and predictions. However, building a good model is
13. Helm V. Principles of Computational Cell Biology: From Protein
a hard task. To help interested readers, we analyzed Complexes to Cellular Networks. Morelnbach, Germany:
in detail the state-of-the-art in modeling. Further- Wiley-VCH, 2007.
more, examples of recent models and applications, at 14. Ho DD, Neumann AU, Perelson AS, et al. Rapid turnover
different scales, are included in the final part of the of plasma virions and CD4 lymphocytes in HIV-1 infection.
Nature 1995;373:123–6.
article.
15. Hofestadt R, Thelen S. Quantitative modeling of biochem-

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


ical networks. Stud HealthTechnol Inform 2011;162: 3–16.
Key Points
16. Hunter PJ, Crampin EJ, Nielsen PMF. Bioinformatics, mul-
 Description of models and their use. tiscale modeling and the IUPS Physiome Project. Brief
 Analysis of the model process: form models objectives to model Bioinform 2008;9(4):333–343.
verification. 17. Younesi E, Kasam V, Hofmann-Apitius M. Direct use of
 Brief description of recent mathematical models in biological information extraction from scientific text for modeling and
systems. simulation in the life sciences. In: 9th International Bielefeld
Conference ‘‘Upgrading the eLibrary: enhanced information
services driven by technology and economics’’ 2009. Bielefeld,
Germany: Bielefeld University Library. dpi:10.1108/
07378830911007637.
References 18. Juty N, Le Novere N, Laibe C. Identifiers.org and
1. Bianca C. Mathematical modelling for keloid formation MIRIAM Registry: Community resources to provide
triggered by virus: Malignant effects and immune system persistent identification. Nucleic Acids Res 2012;40:
competition. Math Models Methods Appl Sci 2011;21: D580–6.
389–419. 19. Karplus WJ. The spectrum of mathematical models. Perspect
2. Bianca C, Pennisi M. The triplex vaccine effects in Comput 1983;3:4–13.
mammary carcinoma: A nonlinear model in tune with 20. Klipp E, Herwig R, Kowald A, et al. Systems Biology in
SimTriplex. Nonlinear Anal Real World Appl 2012;13: Practice: Concepts, Implementation and Application. Darmstadt,
913–1940. Germany: John Wiley & Sons, 2005.
3. Alemani D., Pappalardo F., Pennisi M., Motta S., Brusic V. 21. Lin HH, Ray S, Tongchusak S, et al. Evaluation of MHC
Combining cellular automata and lattice boltzmann method class I peptide binding prediction servers: applications for
to model multiscale avascular tumor growth coupled with vaccine research. BMC Immunol. 2008;9:8.
nutrient diffusion and immune competition. Journal of 22. Lin HH, Zhang GL, Tongchusak S, et al. Evaluation of
Immunological Methods 2012;376(1–2):55–68. MHC-II peptide binding prediction servers: applications
4. Pappalardo F., Motta S., Lollini P.-L., Mastriani E. Analysis for vaccine research. BMC Bioinformatics 2008;9(Suppl 12):
of vaccine’s schedules using models. Cellular Immunology S22.
2006;244(2):137–140. 23. Louzoun Y. The evolution of mathematical immunology.
5. Bianca C., Pennisi M., Motta S., Ragusa M.A. Immune Immunol Rev 2007;216:9–20.
system network and cancer vaccine. International 24. Mitchell H, Levin D, Forrest S, et al. Higher level of repli-
Conference on Numerical Analysis and Applied cation efficiency of 2009 (H1N1) pandemic influenza virus
Mathematics: Numerical Analysis and Applied thanthose of seasonal and avian strains: Kinetics from epi-
Mathematics, ICNAAM 2011, Halkidiki, Greece. AIP thelial cell culture and computational modeling. J Virol
Conference Proceedings, Volume 1389, 945–948. 2011;85:1125–35.
6. Chen P, Pin-Shan P. The Entity-Relationship Model - 25. Mu JC, Jiang H, Kiani A, et al. Fast and accurate read align-
toward a unified view of data. ACM T Database Syst 1976; ment for resequencing. Bioinformatics 2012. doi: 10.1093/
1:9–36. bioinformatics/bts450.
7. Chen LC, Yeh HY, Yeh CY, et al. Identifying co-targets to 26. Palladini A, Nicoletti G, Pappalardo F, et al. In silico mod-
fight drug resistance based on a random walk model. BMC eling and in vivo efficacy of cancer preventive vaccinations.
Syst Biol 2012;6:5. Cancer Res 2010;70:7755–63.
8. Chitnis N, Hardy D, Smith T. A periodically-forced math- 27. Pappalardo F, Pennisi M, Castiglione F, etal. Vaccine proto-
ematical model for the seasonal dynamics of malaria in mos- cols optimization: in silico experiences. Biotechnol Adv 2010;
quitoes. Bull Math Biol 2012;74:1098–1124. 28:82–93.
9. Larman C. Applying Uml and Patterns: An Introduction to Object- 28. Pappalardo F, Forero IM, Pennisi M, et al. SimB16: model-
Oriented Analysis and Design and the Unified Process. New ing induced immune system response against B16-
Jersey, USA: Prentice Hall, 2005. melanoma. PLoS One 2011;6:e26523.
422 Motta and Pappalardo

29. Perelson AS, Neumann AU, Markowitz M, et al. HIV-1 35. Tiger CF, Krause F, Cedersund G, et al. A framework for
dynamics in vivo: virion clearance rate, infected cell lifespan, mapping, visualisation and automatic model creation of
and viral generation time. Science 1996;271:1582–6. signal-transduction networks. Mol Syst Biol 2012;8:578.
30. Lee HY, Topham DJ, Park SY, et al. Simulation and pre- 36. van Iersel MP, Villeger AC, Czauderna T, et al. Software
diction of the adaptive immune response to Influenza A support for SBGN maps: SBGN-ML and LibSBGN.
Virus infection. J Virol 2009;83:7151–65. Bioinformatics 2012;28:2016–21.
31. Polya G. How to Solve It: A NewAspect of Mathematical Method. 37. Zhang GL, Khan AM, Srinivasan KN, et al. Hotspot
Princeton, USA: Princeton Science Library, 2004. Hunter: A computational system for large-scale screening
32. Rumbaugh J, Blaha M, Premerlani W, et al. Object-Oriented and selection of candidate immunological hotspots in patho-
Modeling and Design. New Jersey, USA: Prentice Hall, 1990. gen proteomes. BMC Bioinformatics 2008;9:S19.
33. Segura J, Jones PF, Fernandez-Fuentes N. A holistic in silico 38. Shiwen Zhao S, Li S. A co-module approach for elucidating
approach to predict functional sites in protein structures. drug-disease associations and revealing their molecular basis.
Bioinformatics 2012;28:1845–50. Bioinformatics 2012;28:955–61.
34. Siau K, Halpin T. Unified Modeling Language: Systems Analysis,
Design and Development Issues. Hershey, USA: Idea Group

Downloaded from https://academic.oup.com/bib/article/14/4/411/192938 by guest on 25 October 2024


Publishing, 2001.

You might also like