This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY 1
Robust Calibration of High Dimension
Nonlinear Dynamical Models for
Omics Data: An Application
in Cancer Systems Biology
Fortunato Bianconi , Member, IEEE, Chiara Antonini, Member, IEEE, Lorenzo Tomassoni, Member, IEEE,
and Paolo Valigi , Member, IEEE
Abstract— In computational mathematical modeling of
biological systems, most model parameters, such as initial con-
ditions, kinetics, and scale factors, are usually unknown because
they cannot be directly measured. Therefore, key issues in
system identification of nonlinear systems are model calibration
and identifiability analysis. Currently, existing methodologies
for parameter estimation are divided in two classes: frequentist
and Bayesian methods. The first optimize a cost function, while
the second estimate the posterior distribution of parameters
through different sampling techniques. However, when dealing
with high-dimensional models, these methodologies suffer from
an increasing computational cost due to the important volume of
-omic data necessary to carry out reliable and robust solutions.
Here, we present an innovative Bayesian method, called condi-
tional robust calibration (CRC), for model calibration and identi-
fiability analysis. The algorithm is an iterative procedure based on
a uniform and joint perturbation of the parameter space. At each
step the algorithm returns the probability density functions of all
parameters that progressively shrink toward specific points in the
parameter space. These distributions are estimated on parameter
samples that guarantee a certain level of agreement between each
observable and the corresponding in silico measure. We apply
CRC to a nonlinear high-dimensional ordinary differential
equations model representing the signaling pathway of p38MAPK
in multiple myeloma. The available data set consists of time
courses of proteomic cancerous data. We test CRC performances
in comparison with profile-likelihood and approximate Bayesian
computation sequential Monte Carlo. We obtain a more precise
and robust solution with a reduced computational cost.
Index Terms— Bayesian calibration, cancer systems biology,
nonlinear model, omics data, parameter estimation, system
identification.
I. I NTRODUCTION
B
IOLOGICAL systems are naturally complex, since their
behavior results from the interaction of numerous molec-
ular components [1]. Molecules that govern the basic cell func-
tions are connected in networks and pathways through nonlin-
Manuscript received November 1, 2017; revised March 27, 2018; accepted
May 23, 2018. Manuscript received in final form June 1, 2018. This
work was supported by the Italian Association for Cancer Research under
Grant 15713/2014. Recommended by Associate Editor B. Jayawardhana.
(Corresponding author: Fortunato Bianconi.)
The authors are with the Department of Engineering, University of Perugia,
06132 Perugia, Italy (e-mail: fortunato:bianconi@gmail:com).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TCST.2018.2844362
ear reaction kinetics in which feedback loops play a key role.
Moreover, cells communicate among each other to form higher
organization levels, such as tissues and organs [2]. Thus,
understanding the dynamic interplays in the cell and among
the cells is of crucial importance in order to gain major insights
about mechanisms of complicated diseases [3]. In the context
of systems biology, mathematical, and computational model-
ing are of central importance for reproducing the dynamic
evolution of static pathways and for characterizing biologi-
cal systems from a quantitative point of view [1], [2], [4].
Through control and system identification theory, mathemat-
ical models allow for an abstract and formal representation
of a natural system, while their computational implemen-
tation permits the simulation of the temporal behavior of
model variables, i.e., molecules participating to biochemical
reactions [2], [4], [5]. Although a mathematical model is not an
exact replica model, it can be used for making hypotheses and
predictions on inter or intra-cellular processes [6]. Through
in silico simulations of a model, it is possible to investigate
how a biological process reacts to an external or internal
signal, thus reducing the costs of expensive experiments [7].
The model building process in systems biology is a cycle of
activities organized in the following way. First, it is necessary
to understand the nature of the phenomenon of interest and to
establish the objective and application of the model. Second,
the optimal modeling technique should be chosen in order
to build a first draft model, based also on the a priori
biological knowledge. After that, unknown model parameters
are estimated using the experimental data. Finally, the model
is validated through new experiments, until it reaches the
desired level of agreement with them [4]. Models that pass
this iterative process are declared to be consistent with existing
experimental evidence [8].
As a result, the generation of the experimental data is
strictly related to the modeling process, since it represents
the information that guides both model development and
test [1], [2].
Cancer systems biology represents the application of sys-
tems biology approach to the analysis of how the intracellular
networks of normal cells are perturbed during carcinogen-
esis to develop effective predictive models that can assist
scientists and clinicians in the validations of new targeted
1063-6536 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.