SemGen 3.0 help

Composite annotations

Each composite annotation consists of a physical property term connected to a physical entity or physical process term. The physical entity term can itself also be a composite of ontology terms. We recommend using only terms from the Ontology of Physics for Biology (OPB) for the physical property annotation components. For the physical entity annotations we recommend using robust, thorough, and widely accepted online reference ontologies like the Foundational Model of Anatomy (FMA), Chemical Entities of Biological Interest (ChEBI), and Gene Ontology cellular components (GO-cc). For physical processes annotations, we recommend creating custom terms and defining them by identifying their thermodynamic sources, sinks and mediators from the physical entities in the model.

When you edit a composite annotation for a model codeword, the Annotator provides an interface for rapid searching and retrieval of reference ontology concepts via the BioPortal webservice.

Example: Suppose you are annotating a beta cell glycolysis model that includes a codeword representing glucose concentration in the cytosol of the cell.

A detailed composite annotation would be:

OPB:Chemical concentration <propertyOf> CHEBI:glucose <part_of> FMA:Portion of cytosol <part_of> FMA:Beta cell

In this case we use the term Chemical concentration from the OPB for the physical property part of the annotation, and we compose the physical entity part by linking four concepts - one from the OPB, one from ChEBI and two from the FMA. This example illustrates the post-coordinated nature of the SemSim approach to annotation and how it provides high expressivity for annotating model terms.

The above example represents a very detailed composite annotation, however, such detail may not be necessary to disambiguate concepts in a given model. For example, there may not be any other portions of glucose within the model apart from that in the cytosol. In this case, one could use the first three terms in the composite annotation and still disambiguate the model codeword from the rest of the model's contents:

OPB:Chemical concentration <propertyOf> CHEBI:glucose

Although this annotation approach does not fully capture the biophysical meaning of the model codeword, SemGen is more likely to find semantic overlap between models if they use this shallower annotation style. This is mainly because the SemGen Merger tool currently only recognizes semantic equivalencies; it does not identify semantically similar terms in models that a user wants to integrate. Therefore, if a user wants to integrate our example glycolysis model with a TCA cycle model based on cardiac myocyte metabolism, the shallower approach would likely identify more semantic equivalencies than the more detailed approach.

Nonetheless, we recommend using the more detailed approach, given that future versions of SemGen will include a "Merging Wizard" that will identify and rank codewords that are semantically similar, not just semantically identical.

Extractor

The Extractor tool provides five methods for decomposing SemSim models into sub-models. This decomposition process is useful if you want to "carve out" a smaller portion of a given model in order to remove extraneous model features.

The five extraction methods are represented by the boxes along the left side of the interface: extraction by physical entity, extraction by codeword, and extraction by cluster. They can be used independently or in concert.

Extraction by physical process (SemGen 1.3 and higher): This semantics-based extraction method allows a user to specify which physical processes they want to preserve in the extracted model. In the "Processes" box, select those physical processes in the model you want to preserve and SemGen will retain the codewords annotated with those physical entities in the extracted code. The "include participants" option adds all codewords annotated against the physical entities that participate in the process. In the extracted model, all these preserved codewords are retained along with the input codewords needed for their computation.

Extraction by physical entity: This semantics-based extraction method allows a user to specify which physical entites they want to preserve in the extracted model. In the "Physical entities to extract" box, select those physical entities in the model that you want to preserve, and SemGen will preserve the codewords annotated against those physical entities in the extracted code.

Extraction by sub-model (SemGen 1.3 and higher): SemSim sub-models represent groups of codewords. This extraction method allows a user to specify which sub-models they want to extract. In the "Sub-models" box, select those sub-models in the model that you want to retain, and SemGen will preserve the codewords associated with that sub-model in the extracted code.

Extraction by codeword: This extraction method is included so that users have a maximal level of customization over their extracted model. SemGen preserves any codeword selected in this list in the extracted sub-model, along with the direct inputs needed for its computation. Select the "Include full dependency chain" option to preserve the complete computational dependency map needed to compute a selected codeword. Instead of turning the direct inputs to a selected codeword's computation into user-defined parameters, this option retains the full computational network needed to compute the codeword.

Extraction by computational network cluster: This third extraction method uses a network clustering algorithm to identify the modular portions of a SemSim model's computational network. By identifying the more closely-related computational clusters within a model, this extraction method provides a tool for users who might be unsure about how to best to decompose a model into modular components. Clicking the "Cluster" button in the "Clusters to extract" box will open the cluster identification tool. Raising the slider bar on the cluster identifier iteratively removes edges in the computational network graph of the SemSim model and in turn, delineates higher and higher numbers of computational modules within the model. Sometimes these computational modules will reflect the semantic architecture of the models, and so SemGen also includes a list of physical entities associated with each identified module in a separate window off to the left.

The screenshot below shows a physical entity-based extraction of the left ventricle dynamics from its parent cardiovascular model. Green coloring in the computational network map indicates codewords that are dependent variables in the source model, but will be converted into user-defined inputs in the extracted code.