Premises underlying systems biology research

The focus of research is a biological system. One can broadly define a ‘system’ as a group of independent but interconnected elements that function together to comprise a unified whole. The boundaries of the system may not be clearly defined or definable, especially if multicellular organisms are being investigated. Instead, one may start with an observable phenotype. For example, yeast filamentation, and, using a systems biology approach to identify subsets of modules within a system that specifically relate to the emergence of that phenotype. Systems biologists usually identify a model organism such as yeast, or a model cell population such as macrophages that are well-suited for the biological process under investigation. With a model system in place, systems biology research proceeds using both discovery-based and hypothesis-based approaches.

The hallmark of discovery science is a relatively unbiased collection and cataloging of data pertaining to a specific domain of inquiry. Examples include:

• sequencing of species’ genomes
• non-normalized Expressed Sequenced Tags (EST) libraries of tissue types at various stages of development
• transcriptional profiling and ChIP-chip microarrays
• affinity capture protocols for macromolecular complexes
• delineation of the precise number of cells in an organism as has been done for the worm C. elegans.

Loosely speaking, performing a data collection exercise whose specific results cannot be predicted in advance constitutes a discovery-based approach. Discovery science generates “parts lists” for the system under investigation. And, in some cases, is able to provide a complete and comprehensive list of parts that can be used as a basis for narrowing the scope of possible interactions.

Once a system’s elements and interactions have been discovered and delineated to a first approximation, specific and testable hypotheses are required for determining which elements and interactions are functionally relevant to the observable phenotypes of the system under the varying conditions being investigated. To this end, many systems biologists assume that experimentally observed or inferred interactions among elements might profitably be conceptualized as networks, with the individual elements (e.g., genes, proteins, metabolites) portrayed as nodes, and the interactions or interconnections (e.g., DNA-protein binding, protein-protein binding) as links or edges. In this way, the structure of the system can be conceptualized. Network diagrams provide a visual representation for how different types of interconnections might be organized. (See, for example, Biotapestry and Cytoscape.) These representations are used to guide the formulation of hypotheses.

For example, results of yeast two hybrid experiments performed using different baits and under different physiological conditions can be more easily compared when portrayed as protein interaction networks. From these comparisons, predictions about the content of macromolecular complexes involved with processes, such as gene regulation, can be evaluated as to their likelihood prior to performing expensive experiments.

However, static network models do not reveal dependency relationships among interaction partners or the kinetics of changes in the network over time in response to an experimental perturbation. Thus, they provide a useful starting point for understanding a system but do not, in themselves, explain causal significance (i.e., how functional attributes emerge from the interactions among the network’s components). The distinction between “correlated with” and “functionally affects” is not easily represented in a network diagram of interconnections among a system’s elements. It is anticipated, however, that developments in biological network topology will conduce to more effective modeling tools. Networks cannot easily represent different types of relationships between elements. They cannot indicate that the interaction of elements has changed one of the elements (e.g., chemically modified it), and they cannot represent the dynamics of the network.

The true test of a good system model is successful prediction of the system’s behavior under targeted alterations (genetic or environmental perturbations) of experimental conditions. But the very properties that make biological systems interesting and worthwhile to study their emergent properties, robustness, stability, modularity and adaptability to change, also make their behavior hard to predict at the molecular level. Confounding factors include functional redundancy (i.e., a given process might be accomplished by several different molecular mechanisms), and the stochasticity of cell populations (what is measured, e.g., gene expression, could be an average of a wide range of discrete responses among individual cells).

Systems biologists approach this conundrum by adopting the following principles:

1. Global approaches should be taken to data collection and analyses. Ideally, high-throughput platforms are used to collect accurate measurements under multiple sets of well-defined experimental conditions. Technologies for performing quantitative, multi parameter measurements on a single sample need to be developed. To add value to the analyses of data obtained from multiplex technologies such as chips and panels of gene deletion mutants or RNAi gene knockouts, global approaches will incorporate relevant findings from curated databases and the published literature.
   
   
2. Information derived from diverse data types should be integrated. Systems biology derives power from the leveraging of pre-existing biochemical and cell biology knowledge with the various interaction network models inferred from the global datasets. Even though each source of data type might be sparse, noisy, or contain systematic errors, a meaningful pattern among the diverse data might become apparent and further analysis made possible if the network models are integrated.
   
   
3. Mathematical and statistical modeling is essential to the quantitative analysis of a system’s properties. Based on a working model and relevant assumptions, computer simulations are used to probe the probable effects of perturbations on a system’s components and interactions in the interest of making predictions that can be validated by the collection of more data. Thus, there is a tight integration of computer modeling with experimental design.
   
   
4. Biology should drive technology which, in turn, makes better biology possible. Invention of novel or more sophisticated data collection, analysis and modeling tools is motivated by the need to solve a real-world biological problem. As a paradigm case, the Human Genome Project forced the development of high-throughput DNA sequencing methodologies. The need to perform multiparameter measurements on single cells is currently driving the invention of microfluidic/nanotechnology devices.
   
   
5. Systems biology research should create an interactive inter-disciplinary scientific culture. For progress to occur, experts in engineering, physics, mathematics, and computer science must join biochemists, cell biologists, and physiologists in the effort to figure out how to obtain the required data and develop the sophisticated computational approaches that will be needed to make viable predictions. For scientists who have been trained primarily in one of these disciplines, doing systems biology research involves stepping outside one's comfort zone to learn new concepts and methodologies. Systems biology-focused institutions accept that cross-disciplinary training from the get-go is the best way for new investigators to embrace the field.
   
   
6. The results of research should be freely disseminated. The Human Genome Project has revealed the enormous benefit that derives from the public release of data to the community of researchers. While not as easy to work with as genomic sequence, available microarray datasets, yeast two-hybrid analyses, collections of gene knockout strains and the like have accelerated progress in systems biology research. Similarly, computational biology is facilitated by the sharing of open-source software.