Model alignment with semantic propagation

Click for help
To align two SBML models, model elements are individually compared by
their semantic annotations. Each possible element pair obtains a similarity
value and a greedy algorithm is used to match element pairs with high
similarities.
A model alignment based on MIRIAM-compliant element annotations
provides good results in many cases. Unfortunately, the alignment is
difficult or impossible if one or both of the models
- are missing annotations, which makes it impossible to align some of the elements,
- or contain ambiguous annotations, leaving many different element matches equally probable.
In such cases, model alignment can be improved by using the semantic
annotations of other elements that are nearby in the model network. For
instance, if two reactions from different models share reactants (e.g., if
the reactants have already been matched by their semantic annotations),
they can be regarded as similar. In this case, knowledge about
similarity is propagated from the reactant elements to the reaction elements.
The model alignment on this page lets semantic information diffuse
across the reaction network to infer similarities between
non-annotated elements. There are two alternative methods. In the first
method ("feature propagation"), information about single-element
annotations diffuses on the reaction network. In the second
method ("similarity propagation"), similarities of element pairs are
propagated on an abstract pair matching graph. Differences between the
two approaches and the normal similarity measure can be evaluated with
this tool. After uploading two models, the model alignment will be shown
in the form of a an element pair matrix:
The first row shows pairwise similarities obtained
without propagation,
with
feature propagation, and with
similarity propagation. Color
intensities depict the similarity values. The second row shows the
matching based on these similarity values (color: matched elements;
white: no match). For more details about semantic propagation and
model alignment, please read the upcoming article:
Schulz et al.: Propagating semantic information in biochemical network models.