Graphical Modeling Meets Systems Pharmacology

Abstract

A main source of failures in systems projects (including systems pharmacology) is poor communication level and different expectations among the stakeholders. A common and not ambiguous language that is naturally comprehensible by all the involved players is a boost to success. We present bStyle, a modeling tool that adopts a graphical language close enough to cartoons to be a common media to exchange ideas and data and that it is at the same time formal enough to enable modeling, analysis, and dynamic simulations of a system. Data analysis and simulation integrated in the same application are fundamental to understand the mechanisms of actions of drugs: a core aspect of systems pharmacology.

Keywords: Graphical modeling; data analysis; design tools and techniques; modeling and simulation; stochastic and hybrid simulation algorithms; user-centered design.

Figure 1.

The 3 arrows on the top-right corner of the figure represent, respectively, modifications (actions of a molecule on another, such as phosphorylation and ubiquitination, with positive results), inhibitions (negative effect of a molecule on another), and translocations (movement of a molecule from one compartment to another). An additional dashed line arrow is available for adding narrative clues with no specific biomolecular counterpart. The symbols at the bottom of the figure represent from left to right small molecules (eg, metabolites or drugs), proteins, gene-related material (eg, DNA, RNA, messenger RNA, and microRNA), raw material to represent de novo synthesis and degradation of components, and generic node to model whatever cannot be modeled with the previous node types.

Figure 2.

The visible model is an excerpt of the KEGG (Kyoto Encyclopedia of Genes and Genomes) FoxO signaling pathway in humans as appears after importing in bStyle. The 2 screenshots show the parameterizations for 2 different selected objects that bStyle highlights with an orange border. (A) Transforming growth factor receptor (TGF-βR) on the cytomembrane is selected and the toolbar presents its configurable details. New subclassification user interface levels are available from the toolbar to fulfill user’s need for fine-grained properties. In this case, the click sequence of the user was “Modifications” in toolbar, “Modification” in the selection list, and then “Phosphorylation” from a complete list of possible covalent protein modification statuses. (B) The TGF-β signaling pathway, that KEGG simplifies to a horizontal cascade arrow, is presented accordingly in the model and in its counterpart toolbar representation, showing the intervention of SMAD4 to result in SMAD3.

Figure 3.

The gemcitabine cascades described in Garcĺa-Manteiga et al., here modeled with bStyle. The user has selected the phosphorylation by deoxycytidine kinase (dCK) of gemcitabine (dFdC) to its monophosphate dFdC-MP. The entire transformation is presented in the toolbar with a familiar symbolism. Additional nonuser-selected entities are presented as well to provide the most complete representation of the modeled process. In this case, the competition of gemcitabine with the natural nucleoside deoxycytidine diphosphate (dCTP) modeled as the inhibition on dCK via binding with dCTP is clear. Link numbers in the toolbar stand for the corresponding stoichiometric coefficients that can be edited with a click as happens for the rates.

Figure 4.

Tabular view of the gemcitabine model, whose inner compartments have been expanded to show the hierarchical structure. All elements contain active links that allow, in a customary tabular format, to easily inspect or edit any detail of the model. The tree can be filtered via instant search (see also Figure 6) or predefined filters as “Show All,” “Elements only,” or “Reactions only.” The parameterization happens through the same user interfaces seen in the visual model for the best usage experience. Besides the predefined columns “Model Elements” and “Properties,” additional columns can be added manually via user interface or by importing data files. For example, time-series data have been added here as data columns timepoint_4, timepoint_12, and timepoint_16.

Figure 5.

Besides attribute columns of common types as numeric, texts, or series thereof, it is possible to enrich model elements with annotations referring external hierarchically organized ontologies. The Systems Biology Ontology annotation process is shown here, typically used within SBML models and natively supported even when working with non-SBML models. In contrast to other attribute types, the nested structure of IDs with their corresponding descriptions is preserved in the dropdown lists.

Figure 6.

When many attributes and annotations are added to the model, bStyle offers a comfortable way of accessing them all from the visual model. Clicking on the “Annotation” link, a compartmentalized view of the selected element is presented along with all its attributes and annotations, if any, as if it was seen on the tabular view. In complex models, this verticalization is helpful together with the visual model not only by instantly showing the full hierarchy of the selected model but also by providing all associated data right at one’s fingertip. Note how the user changed the color of selected interaction arrows to facilitate the visual aggregation and identification of specific subprocesses of interest.

Figure 7.

When the modeling complexity increases, spotlight search comes into play to help quickly spot specific elements in crowded models. By autosuggesting search results, the feature assists the user in picking the desired biochemical name. In addition, the visual spotlight highlights the elements identified by the typed characters. The spotlight search view can be retained and saved for publication as well. A “spotlight view” is also available: from the entire model in the shadow, the user can manually turn on the spotlight on the desired elements (with a click). The spotlight mode of the elements is retained internally and can be iteratively updated along time. Spotlights and spotlight searches allow dramatic model picks that can be saved or sent to print.

Figure 8.

The simulation toolkit is organized into 2 main tabs, the “Simulation” tab and the “Timeseries” tab. (A) An extensible set of simulation algorithms can be hosted within the rational interface of bStyle. Besides the built-in RSSA, HRSSA, DM, Euler, and RK45 algorithms, a COmplex PAthway SImulator simulation bridge is available to instruct their simulation engines after the currently open model. When the same simulation parameter represents the same concept across different algorithms, then it is presented at the same location on screen allowing to compare different algorithms on the same parameters. Common for all algorithms is the possibility to preselect, before running the simulation, a subset of the model either through a direct selection from the hierarchical element list or using only the selected or spotlighted elements (if any). (B) The simulated model is graphically presented through the time series of its elements. A compartment filter is available along with a rich set of useful graphical options to analyze the results. DM indicates direct method; HRSSA, hybrid rejection-based stochastic simulation algorithm; RSSA, rejection-based stochastic simulation algorithm.

Figure 9.

Interoperability with external services is naturally available with uniform interfaces harmonizing possible differences across different providers. Therefore, opening a pathway map or the creation of a geneset or an edgeset shows up with a very similar aspect with clear differences to differentiate the different processes. (A) Opening a remote pathway, it is just a matter of selecting the remote database (DB) and then picking the requested pathway with the customary autosuggest search boxes and DB-specific features. (B) For generating genesets and edgesets, the user can select the DB, the kind of molecules to export from the available ones in the selected DB, and save to standard GMT format for further external processing.

Figure 10.

Using external knowledge on the biological localization of the elements, bStyle was used to graphically render the identified pathways with different sizes and colors based on the value of their Network Activity Score. This resulted in the first global view of the active processes in human adipogenesis at the different time points and across time lapses.