ModelDB: a Tool for Model Sharing
ModelDB is a curated database of published models in the broad domain of computational neuroscience.
It addresses the need for access to such models in order to evaluate their validity and extend their use.
It can handle computational models expressed in any textual form, including procedural or declarative languages (e.g. C++, XML dialects) and source code written for any simulation environment.
The model source code doesn't even have to reside inside ModelDB; it just has to be available from some publicly accessible online repository or WWW site.
ModelDB is curated in order to maximize the scientific utility of its contents.
The ideal model entry would contain "original" (author-written) source code, especially if it works and reproduces at least one figure from a published article.
Original source code has tremendous value because it is what the authors used to generate the simulation results from which they derived their published insights and conclusions.
High quality "third party" re-implementations of published models are also relevant, especially those involving models that are of wide interest.
ModelDB's historical focus on computational models does not deny the importance of clear and complete descriptions of conceptual models ("hypotheses").
For that matter, there is no reason in principle why detailed descriptions of conceptual models could not be stored in ModelDB.
Indeed, this is likely to happen spontaneously as the distinction between conceptual and computational specifications is blurred by the gradual evolution of model specification styles from procedural to declarative.
That said, it is important to note again that simulation results are generated from computational models, not from conceptual models.
The correctness and completeness of any description of a conceptual model, and the supposed match between that model and its expression in computational form, are only assertions that remain to be proven, regardless of the form in which the conceptual model is stated--verbal summary, equations, tables, charts, diagrams, or some combination of these.
From the perspective of model authors and readers alike, it is of paramount importance to be able to verify the match between conceptual model and its computational implementation.
Only by having access to both the conceptual model and the authors' source code can readers verify this for themselves.
Benefits of ModelDB
ModelDB contains a large and growing collection of published models that are searchable and downloadable.
This makes it of immense potential value to a wide range of individuals and groups.
For model authors in particular
- improves scientific productivity by making it easier to discover published models
- serves as a source of potentially reusable elements e.g. parameter sets, architectures of cells and networks, mathematical descriptions of biophysical mechanisms
- facilitates communication of ideas to colleagues
- stimulates attributed re-use of published work
For developers of simulation environments
- provides models for regression testing, to insure correctness and compatibility of new versions with prior releases
For developers of tools for model exchange
- provides models for testing translators and validating strategies for expressing models in interchangeable formats
For the field of neuroscience as a whole, and computational neuroscience in particular
- catalyzes advances in computational neuroscience research
- enhances impact of computational modeling on scientific progress
For publishers of scientific journals
- promotes citation of modeling articles
Source code availability is essential for the evaluation
of published computational models
Here are just two of the many reasons why source code availability is essential for the evaluation of published computational models.
It can help provide a clear picture of what was actually studied.
Descriptions of conceptual and computational models in published papers are generally ambiguous or incomplete or both.
Source code can compensate for such deficiencies.
It is indispensable for deciding whether a particular paper satisfies "the fundamental premise of computational modeling," i.e. the assumption that there is a close match between a computational model and the conceptual model for which it is to serve as a surrogate.
The ability to verify such a match is of paramount importance to model authors and readers alike.
Only by having access to authors' source code can readers do this for themselves.