RuleBender: integrated modeling, simulation and visualization for rule-based intracellular biochemistry

Abstract

Background: Rule-based modeling (RBM) is a powerful and increasingly popular approach to modeling cell signaling networks. However, novel visual tools are needed in order to make RBM accessible to a broad range of users, to make specification of models less error prone, and to improve workflows.

Results: We introduce RuleBender, a novel visualization system for the integrated visualization, modeling and simulation of rule-based intracellular biochemistry. We present the user requirements, visual paradigms, algorithms and design decisions behind RuleBender, with emphasis on visual global/local model exploration and integrated execution of simulations. The support of RBM creation, debugging, and interactive visualization expedites the RBM learning process and reduces model construction time; while built-in model simulation and results with multiple linked views streamline the execution and analysis of newly created models and generated networks.

Conclusion: RuleBender has been adopted as both an educational and a research tool and is available as a free open source tool at http://www.rulebender.org. A development cycle that includes close interaction with expert users allows RuleBender to better serve the needs of the systems biology community.

Figure 1

The RuleBender interface. Shown are the Model Editor pane including console for text output (left) and the Visualization Viewer pane (right). The Visualization Viewer shows two complementary visual encodings corresponding to the text model in the Editor: the interactive contact map (top), and part of the influence graph for this model (bottom). RuleBender's main features include syntax checking, syntax highlighting, visual global model exploration with linked views, integrated execution, support for multiple simulation modules, simulation journaling, interactive plotting including comparison of multiple datasets, and parameter scanning.

Figure 2

The Contact Map. Contact Maps without (left) and with (right) hub nodes. Molecules are represented as larger nodes (light gray) while domains and domain states (yellow, orange and purple) are represented as smaller sub-nodes in the molecules. State nodes (green and dark gray) are adjacent to the domain sites to which they apply. Reaction rules are mapped to edges (rules that indicate the creation or destruction of a bond between these two domains) and state nodes (rules that indicate state changes). Selecting a state-node (red boundary on the left) lists all rules that indicate that state change. Similarly, selecting an edge (not shown) lists all rules that create or destroy bonds between the linked domains. Selecting one rule from such a list marks the reaction context in blue and the reaction center in pink. Hub nodes are associated with rules that define molecular level interactions without domains involved, such as the degradation of proteins. Selecting a hub node lists all rules involving the linked molecules as shown on the right.

Figure 3

Compartmental Contact Map. Contact Map with molecule compartment hierarchy (extracellular, cytoplasmic, nucleus etc). The saturation of the convex hull encompassing a compartment indicates the hierarchical structure of the compartments; the outermost compartment is colored the lightest blue. All the members of a compartment can be moved as a whole unit to get a clear view of the hierarchical structure.

Figure 4

The influence graph. Nodes represent reaction rules while arcs represent influence between rules. Green/Magenta solid arcs represent fully activation/inhibition, and Green/Magenta dash arcs represent partial activation/inhibition. Filter options that show or hide activation/inhibition are provided through pop-up menus. Two separate groups of rule nodes (group1: the first four nodes, group2: the rest of the nodes). can indicate that the model is not complete.

Figure 5

Influence graph definition. Prototype pattern relations (P) and rule relations (R) used to determine influence graphs: an intermediate graph (Left) is ultimately reduced to the simplified, final influence graph (Right). An arrow from P to R means that P is a reactant pattern of the R; for the reverse direction P is a product pattern of R. Green edges show activation relations and red ones show inhibition.

Figure 6

The simulation results viewer. The upper left quadrant of the window contains a file explorer for easy retrieval of the exact version of a particular model associated with a specific set of results; the bottom left quadrant shows the list of species or observables. Charts in linear or log scale show the time series for concentrations of chemical species and observables. Any number of species and observables can be compared in the same chart. Furthermore, multiple simulation runs can be compared in order to analyze the effects of changing the model. The example in the snapshot compares the results of two simulations (points and lines) with three observables selected individually.

Figure 7

The species graph. The species graph is constructed similarly to the Contact Map. Shown is an example of a complex species containing thirteen molecules which is difficult to grasp from the text representation only.

Figure 8

Lyn-binding Contact Map. Contact Map visualization for the Lyn-Binding model.

Figure 9

Lyn-binding debugging. Reduced view: Ligand notation shortened to L and Rec shortened to R. If the user programs the rule that binds Lyn to Rec incorrectly (see Table 2), the corresponding contact map in (a) is missing the rule context information. The correct binding leads instead to the visualization in (b); the presence of the blue bubble set alerted the researcher to the difference and allowed them to debug their RBM. The incorrect formulation would allow at run time for the creation of the infinitely binding chain shown in (c).