Theory, Modeling, and Simulation Coupled to Experimentation of Complex Biological Systems
Theory, Modeling, and Simulation Roadmap
Theory and Modeling Objective: Build concepts and models of microbial cells and communities that capture and extend our knowledge, based on a combination of experimental data types.
Simulation Objective: Test and validate concepts and use integrated models to understand mechanisms, explore new hypotheses or conditions, and drive new experimentation.
The only conceivable methodology for success in achieving GTL goals is a coherent and tight integration of theory, modeling, and simulation (TMS) with experimentation (E) and resultant data. Theory refers to the hypothetical concept that underlies properties and phenomenological behavior. Modeling is the translation of that theoretical concept into mathematical terms so calculations can be carried out. Simulation combines multiple models into a meaningful representation of the whole system, encompassing physicochemical and other variables that together evolve computationally to identify “emergent” behaviors.
Computationally driven TMSE provides an interface between the researcher and huge resultant data sets from complex systems, involving (1) at a mechanistic level, multiple strongly interacting processes and elements; (2) at a functional level, multiple strongly coupled phenomena; and (3) behaviors that are unforeseen and not intuitively accessible. Rapid and inexpensive in silico experiments via simulations can be used to gain first insights, form hypotheses, and conceive and carry out meaningful tests. Utilizing simulations for understanding critical parameters, investigators can technically and statistically design physical experiments for maximum efficacy. Resultant data from all experiments will be compared against simulations in various ways to test assumptions and hypotheses, identify new phenomena, and spark new theories. Computational simulations are a time machine, microscope, and telescope, allowing complex systems to be analyzed from any conceivable organizational, temporal, spatial, process, or phenomenological perspective.
To address DOE’s mission interests, we will need to go beyond understanding how cells work in known environments. We must predict how organisms will respond to new sets of conditions, how selected collections of components might be put to work in vitro (another set of conditions), and how we can tune the biochemical processes to do different things. In other words, a chief goal in making models and simulations will be to apply them to circumstances different from the situations for which we have data.
When dealing with complex systems, TMS and high-performance computing have emerged as the universal methodology to drive experimentation, which can be prohibitively expensive, difficult, and time consuming. The use of TMS has become the foundational capability for every aspect of science and engineering, from chemical engineering to aerospace designs, and has fostered a dramatic change in research and technology-development cycles. The GTL Knowledgebase will serve as a discovery-driven resource for developing new modeling capabilities, conducting experiments to verify models quantitatively, and simulating how interacting phenomena affect each other. Insights gained will be readily accessible in the GTL Knowledgebase for new hypotheses and statistically designed experiments, bringing to bear cumulative, cross-referenced data on building new models and simulations.
Ultimately, scientists will be able to create in silico models of a microbe by comparing its genomic sequence with highly annotated gene and functional information in the knowledgebase. The goal is to create increasingly accurate mathematical models of life processes to enable prediction of cell and community behavior and develop new or modified systems tailored for mission applications (see Examples of Biological Understanding and Possible Applications Enabled by TMS).
Microbial Behavior: Modeling at the Molecular Level
The starting point for GTL analysis is to decipher microbial processes at the molecular level. The centerpiece of GTL is the ability to analyze, reconstruct, and model the networks of molecular interactions at the core of life processes. Cell networks arise from the series or chains of molecular interactions during metabolism, protein synthesis and degradation, regulation of genetic processes such as transcription and replication, and cell signaling and sensing. In short, cellular molecular networks and pathways are at the center of cell modeling and cellular behavior and, ultimately, of microbial-community modeling and behavior. Such models would predict how a cell’s genome and environmental factors combine to yield its phenotype. Models will be powerful tools for scientific discovery as we explore the enormous complexity of microbes and their communities.
Computer Science and Mathematics Challenges
Achieving predictive capabilities will require overcoming many technical challenges. For example, cell modeling eventually may involve a more complex collection of components and materials than do existing models of climate or mechanical systems. Many needed developments involve research in computer science and mathematics. New mathematical methods are needed for analysis of raw biological data to include in models and the subsequent statistical design of experiments to validate those models. Additionally, major research challenges relate to database query and design in support of modeling, as well as the development of effective databases to capture modeling output and the models themselves.
Modeling complex biological systems will require new methods to treat the vastly disparate length and time scales of individual molecules, molecular complexes, metabolic and signaling pathways, functional subsystems, individual cells, and, ultimately, interacting organisms and ecosystems. Such systems act on time scales ranging from microseconds to thousands of years. These enormously complex and heterogeneous full-scale simulations will require not only petaflop capabilities but also a computational infrastructure that permits model integration. Simultaneously, it must couple to huge databases created by an ever-increasing number of high-throughput experiments. Challenges include determining the right calculus to describe regulation, metabolism, protein interaction networks, and signaling in a way that allows quantitative prediction. Possible solutions include use of differential equations, stochastic or deterministic methods, control theory or ad hoc mathematical network solutions, binary or discrete value networks, Chaos theory, and emerging and future new abstractions.
Fundamental Questions and Issues
As this systems-level approach to understanding microbial cells and their communities develops, several questions must be addressed:
- What are the biological design variables?
- Can biological systems be modeled to the same degree as physical and chemical systems?
- How do physical and chemical principles and approximations developed for modeling nonliving systems apply to the simulation of living systems?
- Are numerical values for parameters such as enzyme-catalyzed reaction rates known, or even knowable, since such properties change with time and environmental conditions and from cell to cell?
- How can we quantify the levels of uncertainty in our understanding and predictions and the sensitivity of our models to variations in input parameters and structure?
- How do we address important issues of model and knowledge representation and formalism?
Chemistry Challenges
Chemistry is essential to our understanding and exploitation of cellular processes. The functions of a cell are increasingly being understood through explanation of the underlying chemistry. Structural-imaging technologies enable construction of models of protein machines as they carry out many cell functions, including processes relating directly to DOE missions that are the focus of GTL. As we learn more about cells, we will want to broaden the range of operating conditions to which these protein machines can be exposed and the range of environmental substrates that they can convert to other substances. We also will modify proteins to make them active with non-native substrates of interest to DOE, resulting in levels of specificity different from the native system. For example, a machine that enables a microbe to convert an environmental contaminant such as carbon tetrachloride to benign products might be modified to enable the microbe to destroy the related contaminant trichloroethylene. Understanding the detailed chemical mechanisms taking place as the protein machine processes a substrate will be critical to planning its use intelligently and engineering it to meet our needs.
Structure, Interactions, and Function
Reliable high-throughput determination of protein and protein-complex structures and functions will require computational methods capable of integrating several sources of experimental data; examples include mass spectrometry (MS), X-ray crystallography and scattering, protein arrays, numerous imaging modalities, cross-linking, yeast two-hybrid, and nuclear magnetic resonance (NMR). High-throughput MS experiments involving complexes and cross-linkers pose significant informatics and computational challenges.
These data sets will enable molecular-level simulations and prediction, thus populating the GTL Knowledgebase with functional annotations at three levels: (1) Computationally driven high-throughput protein-structure prediction, (2) integrated experimental and computational approaches to structures and function for hard-to-isolate proteins and complexes, and (3) advanced molecular simulations of biochemical activity.
An important driver for high-performance computing systems will be modeling and simulation to predict the behavior of complexes of specific protein sets chosen from network analyses and other experiments. Computational requirements for such simulations are the best characterized among all areas of computational biology; moreover, many of these simulation methods already are implemented on teraflop-scale computers. Pure computing power is the major limitation on size and accuracy of many biochemical simulations, which will involve data and models of protein-protein interactions, ligand-protein interactions, electron-transfer interactions, and membrane characteristics. Molecular dynamics and quantum mechanics–based molecular modeling will spur high-end computing and require development of more-effective scalable algorithms. The GTL program will push the envelope for biophysical modeling, in particular, to develop the ability to predict the actual behavior of proteins and protein complexes for a set of biological processes chosen for their importance to GTL goals.
Microbial Behavior: Metabolic Network and Kinetic Models of Biochemical Pathways
Current State of Cell-Network Modeling: Moving from Experiment (Real Life) to Simulation (Abstract Systems Model)
The primary goal of cell-network modeling is to capture in an abstract mathematical model the structure (topology), kinetics, and dynamics necessary to analyze and simulate the behavior of networks present in a particular organism. Models are constructed from a combination of mathematical principles and experimental data (e.g., from annotated genomes, proteomics databases, in vitro experiments, expression, and the historical literature). Models are used to facilitate a general understanding of cellular networks and for simulations that attempt to reproduce or predict a particular experimental result. When attempting to develop a systems understanding of complex biology, investigators will use simulation and modeling as one of a few ways to derive insight from complicated interactions involving numbers of variables and details that cannot be grasped intuitively.
Current state-of-the-art models can be used to make specific quantitative predictions for limited regions of well-characterized metabolic pathways or a limited set of specific regulatory or signaling circuits. Although more-general qualitative predictions can be made for larger, more complete networks, the current lack of kinetic constants for most enzymes and of concentration data for intermediate metabolites limits the ability to simulate quantitative results for entire networks, including cells and communities. Modeling also is hampered by the incomplete specification of networks due to lack of functional gene assignments, protein complex and association data, and data for regulatory elements and interactions. Bioinformatic techniques are used upstream of modeling and simulation to extract from experimental data the relationships and functions needed for simulation.
Mathematical-analysis techniques are used to further develop, understand, and improve abstract models and our ability to simulate them. A number of software systems have been developed to model and simulate cell networks (e.g., Gepasi, E-Cell, V-Cell, DBsolve, ChemCell, and BioSpice). Several different formalisms (e.g., rule based, ordinary differential equation, logical, and qualitative) represent and simulate cell-network models. Current cell-network simulations typically are running on serial computers (PCs and workstations) and are used mostly to simulate processes in individual cells or simple cellular interactions.
Advanced Modeling Capabilities
No dominant formalism, however, has emerged that can satisfactorily represent both the kinetics and dynamics of metabolic networks and the logical structure of signaling and regulation. Much new work is needed in this area. Another critical topic that must be addressed is how best to represent multiple levels of spatial and temporal scales in cellular systems and incorporate them into models. Most models of cellular networks are one dimensional (1D) (e.g., box models that assume a completely mixed environment). To make progress toward the ultimate goal of accurate phenotype prediction, future modeling schemes need to incorporate 3D modeling and intracellular compartmentalization. Multiple modeling and inference techniques can address different classes of problems, each with distinct temporal and spatial scales and each with potentially different computational complexity. All classes of problems have specific data limitations and a diverse set of data sources, as mentioned above. Limitations on the models themselves depend on the levels of abstraction used and the mathematical treatment of the problem.
Compartmentalized models will become increasingly important for depicting distinct types and phases of metabolism in organisms such as cyanobacteria, which have both oxygenic and anoxic pathways separated either spatially or temporally. Compartmentalized models will be needed to fully describe life cycles of prokaryotes, which include mechanisms such as sporulation, heterocyst formation, and differentiation. Models with multiple compartments will have to address coupling of compartments (e.g., data and flux representations and stability and fidelity) in a scalable fashion. Much may be learned from the experiences of the DOE National Nuclear Security Agency’s Accelerated Strategic Computing and the climate modeling community. Compartmentalization and coupling also will become an issue in multicellular systems (e.g., bacterial communities and multicellular organisms). A major modeling challenge is the choice and effective exploitation of mathematical abstractions. Biological systems differ from those produced by human engineering in that hierarchies or functional subsystem modules are not necessarily obvious, yet exploiting modularity or lumping the system may be essential for efficient modeling and simulation.
Crosscutting Research and Development Needs
A major challenge is the need to integrate heterogeneous data types into cell models for molecular interactions, metabolism, and regulation. Types include data generated by different imaging modalities, structure determinations, MS, coexpression analyses, and an array of binding and other constraints.
Mathematical models ultimately must be developed from fundamental biological principles. Mathematics and computer science research will aid in understanding the following:
- Organization of principles or theories that could lead to successful models, even with incomplete knowledge, missing data, and errors.
- Determination of strengths and weaknesses of different types of simulation methods for different systems biology problems (e.g., stochastic, differential equation, mathematical networks).
- Use of high-performance computing to provide the compute power to run long time-scale simulations (e.g., in milliseconds and longer time frames for ab initio or directed and constrained molecular dynamics for simulating machines).
Computationally, no single architecture is appropriate for all aspects of predictive cell modeling. Because computational requirements are so diverse, coupling informatics with modeling and simulation establishes the need for a fully general-purpose computing infrastructure. Hardware needs for such a challenge range from commodity clusters to tightly coupled, massively parallel architectures with greater investment in interprocessor communication. Implications for operating systems are equally disparate, requiring in some cases extremely high rates of parallel input-output to move data among processors and memory, as well as efficient management of single-application codes distributed over hundreds or thousands of processors.
