Filling in the gaps in cell signaling pathway models
In previous posts on The Digital Biologist, I have discussed the important issue of using incomplete models. This is one of the challenges that I feel has dogged the wider acceptance of modeling in mainstream research in biology. The majority of life scientists who typically get little or no exposure to modeling in their training and subsequent career experience, often ignore modeling approaches in their own research because biological models at the cellular level are largely incomplete and a model that is not complete is not useful - right? It is my hope that in these previous discussions, I have been able to convince at least a few of you that this assumption is not correct and that models can have significant value beyond the narrow confines of the kind of "turnkey" predictive applications that many scientists assume is all they are good for.
All of this is not to say however, that being able to improve and expand a model by filling in the gaps where data and knowledge are missing, is not a desirable goal.
In a recent PLOS One paper by a group at Carnegie-Mellon University, a graph-theoretical approach is presented by which missing interactions in cell signaling pathways can be predicted based upon the shortest path problem - i.e. the determination of the path between the vertices of a weighted graph, for which the sum of the weights of its traversed edges is a minimum.
In essence, the authors' approach involves searching for new nodes in the graph (corresponding to new protein interactions in the pathway) that could substantially shorten the path from the upstream signaling proteins (that the authors call "sources") to the downstream effectors ("targets"). In testing their approach, they observe that in general it is possible to dramatically shorten the paths through incomplete signaling cascade graphs, with only a few additional edges - furthermore, they observe that these putative new interactions are generally supported by peripheral evidence in the existing literature and in the protein interaction databases.
The full text of the paper is available online at the NCBI.
© The Digital Biologist | All Rights Reserved