Processes 2019, 7,37
predictions can be made for the behavior of the signaling system in a qualitative, semi-quantitative
or quantitative manner. For example, for quantitative predictions of signaling and regulatory gene
networks, continuous variables need to be modeled on continuous time scales using ODEs [151].
Such detailed modeling requires knowledge of important biological reactions at every step.
This can only be achieved in several iterative steps that include the implementation of various
components such as signaling events and defining values for related parameters and appropriate
initial conditions. In recent years, numerous models have been developed that simulate individual
signaling pathways [152–154]. The challenge is that these models often do not fully capture crosstalk
mechanisms that are crucial in predicting patient response to treatment, as each drug perturbs multiple
biological processes. ODE-based models can be combined with agent-based models to capture the
dynamics of the system being modeled in a more complex fashion. Stochasticity is one of the main
advantages of agent-based models, as it applies to biological processes [155]. In comparison to ODE
models that make predictions of concentrations and other events over time, ABM allows the study
of each agent, as it interacts with other agents in their proximity and the ways in which that affects
the large-scale behavior. These models, however, are computationally more expensive. There are
also challenges in validating results from ABM due to insufficient spatio-temporal data on tumor
development [156].
One aspect that is typically missing from mechanistic knowledge-based models including QSP
and ABM is an input from high-throughput data, genomic or proteomic; such data can inform the
models and can supplement the data obtained at the cellular and tissue levels [157]. Examples include
immune landscape information from different sources including patients’ databases such as TCGA
(The Cancer Genome Atlas) [33,158,159]. Another source of “Big Data” for parameterization and
validation of the models, including ABM, will be the emerging methodologies of digital pathology,
such as using multiplex immunofluorescence (mIF) of patients’ biopsies and resected tumors, with
subsequent analysis of cellular and molecular spatial patterns. Steps in this direction are already
underway [160]. Another source of data is image-based, using microCT, confocal, multiphoton,
and super resolution microscopy, both ex vivo and in vivo; examples include imaging entire tumor
vasculature with subsequent computer simulation of blood flow and molecular transport [161,162].
In addition to modeling approaches used to simulate response to treatment, virtual patients are
a key component of virtual clinical trials. A virtual population has the characteristics of the original
patient population but also includes individual diversity, usually comprising parameter sets weighted
by a clinical or response distribution [163]. This diversity allows testing a broad range of responses
that can be missed in a clinical trial. In contrast to traditional clinical trials that can only be performed
after costly and lengthy development, in silico trials could be performed at every stage of the drug
development. In silico design of treatments can be conducted with data-driven or mechanistic-based
(knowledge-based) approaches [164]. It should be noted that in silico clinical trials require integration
of data at different scales via a multi-model approach using virtual patients. Similar to a traditional
clinical trial, rigorous statistical approaches are needed at various steps of virtual clinical trials. The
ability to test treatment via in silico clinical trials can significantly reduce the cost and increase the
efficacy of drug development.
The main challenge in the way of predictive models for virtual clinical trials is the availability of
input data for the model for each patient. Detailed knowledge about the situation at the start of the
simulation can significantly affect the predictive power of that model. Such input information is being
generated at a growing rate and a lower cost. Furthermore, proteomic data enable the modeling of
interactions of different subgroups of cells from the same tumor with each other as well as immune
cells and other stromal cells, allowing modeling tumors for individual patients. Computational models
can make predictions for the optimal treatments, making it safer, faster, and cheaper to complement
current clinical trials. These models will improve by continuous comparison of predicted and actual
response to therapy. Additionally, as more detailed information on biological parameters and disease
mechanisms become available, the accuracy of the models will increase.
54
predictions can be made for the behavior of the signaling system in a qualitative, semi-quantitative
or quantitative manner. For example, for quantitative predictions of signaling and regulatory gene
networks, continuous variables need to be modeled on continuous time scales using ODEs [151].
Such detailed modeling requires knowledge of important biological reactions at every step.
This can only be achieved in several iterative steps that include the implementation of various
components such as signaling events and defining values for related parameters and appropriate
initial conditions. In recent years, numerous models have been developed that simulate individual
signaling pathways [152–154]. The challenge is that these models often do not fully capture crosstalk
mechanisms that are crucial in predicting patient response to treatment, as each drug perturbs multiple
biological processes. ODE-based models can be combined with agent-based models to capture the
dynamics of the system being modeled in a more complex fashion. Stochasticity is one of the main
advantages of agent-based models, as it applies to biological processes [155]. In comparison to ODE
models that make predictions of concentrations and other events over time, ABM allows the study
of each agent, as it interacts with other agents in their proximity and the ways in which that affects
the large-scale behavior. These models, however, are computationally more expensive. There are
also challenges in validating results from ABM due to insufficient spatio-temporal data on tumor
development [156].
One aspect that is typically missing from mechanistic knowledge-based models including QSP
and ABM is an input from high-throughput data, genomic or proteomic; such data can inform the
models and can supplement the data obtained at the cellular and tissue levels [157]. Examples include
immune landscape information from different sources including patients’ databases such as TCGA
(The Cancer Genome Atlas) [33,158,159]. Another source of “Big Data” for parameterization and
validation of the models, including ABM, will be the emerging methodologies of digital pathology,
such as using multiplex immunofluorescence (mIF) of patients’ biopsies and resected tumors, with
subsequent analysis of cellular and molecular spatial patterns. Steps in this direction are already
underway [160]. Another source of data is image-based, using microCT, confocal, multiphoton,
and super resolution microscopy, both ex vivo and in vivo; examples include imaging entire tumor
vasculature with subsequent computer simulation of blood flow and molecular transport [161,162].
In addition to modeling approaches used to simulate response to treatment, virtual patients are
a key component of virtual clinical trials. A virtual population has the characteristics of the original
patient population but also includes individual diversity, usually comprising parameter sets weighted
by a clinical or response distribution [163]. This diversity allows testing a broad range of responses
that can be missed in a clinical trial. In contrast to traditional clinical trials that can only be performed
after costly and lengthy development, in silico trials could be performed at every stage of the drug
development. In silico design of treatments can be conducted with data-driven or mechanistic-based
(knowledge-based) approaches [164]. It should be noted that in silico clinical trials require integration
of data at different scales via a multi-model approach using virtual patients. Similar to a traditional
clinical trial, rigorous statistical approaches are needed at various steps of virtual clinical trials. The
ability to test treatment via in silico clinical trials can significantly reduce the cost and increase the
efficacy of drug development.
The main challenge in the way of predictive models for virtual clinical trials is the availability of
input data for the model for each patient. Detailed knowledge about the situation at the start of the
simulation can significantly affect the predictive power of that model. Such input information is being
generated at a growing rate and a lower cost. Furthermore, proteomic data enable the modeling of
interactions of different subgroups of cells from the same tumor with each other as well as immune
cells and other stromal cells, allowing modeling tumors for individual patients. Computational models
can make predictions for the optimal treatments, making it safer, faster, and cheaper to complement
current clinical trials. These models will improve by continuous comparison of predicted and actual
response to therapy. Additionally, as more detailed information on biological parameters and disease
mechanisms become available, the accuracy of the models will increase.
54