Doing systems biology: workflow, data and tools

Systems biology follows a cycle:

• Define the problem.
• Design and run experiments.
• Annotate and curate the data, checking it against published research and databases.
• Infer the network of interactions.
• Build and simulate a model.
• Compare its predictions with experiments.
• Update the hypothesis and the model, and go round again.

If the model predicts a protein level twice as high as measured, that disagreement is useful: it shows what to change next. A big model can seem frightening at first, so break it into smaller parts.

Managing the data has three aspects: minimum information (what must always be recorded about an experiment); standard file formats, based on XML (just as your phone saves photos as JPG and documents as PDF, so that anyone can open them); and ontologies, agreed vocabularies such as GO (Gene Ontology) and SBO (Systems Biology Ontology). Data are kept in spreadsheets, electronic lab notebooks (ELN) and lab management systems (LIMS). Workflow systems such as KNIME, Taverna and Galaxy link the tools together so analyses can be built and shared.

Tools for each job (they keep being updated, like the apps on your phone): data management (Taverna, MAGE-TAB); network inference (MATLAB, R, BANJO); curation (CellDesigner, PathVisio); simulation (JSim, ANSYS); molecular interactions (AutoDock Vina, GOLD); and physiological modelling (PhysioDesigner, OpenCell, FLAME).

Many models are sets of partial differential equations, which usually cannot be solved exactly. They are solved approximately by the finite element method, which breaks the problem into many small pieces (using ANSYS, FreeFEM++ or MATLAB).

Once built, a model is analysed:

• Sensitivity analysis: how the model responds to small changes in its parameters, showing how stable and controllable the system is (SBML-SAT).
• Bifurcation and phase-space analysis: finding the system's steady states and how it changes over time (XPPAut).
• Metabolic control analysis: how much each reaction controls a whole network at steady state (MetNetMaker).