Following in the footsteps of the microarray and proteomics communities, a team of biochemical modeling experts has proposed a set of guidelines that aims to standardize quantitative, machine-readable models of biological systems.
The proposed standard, called MIRIAM (Minimum Information Requested in the Annotation of Models), parallels similar efforts like MIAME (Minimum Information About a Microarray Experiment) and MIAPE (Minimum Information about a Proteomics Experiment), which have proved to be successful in other biological research domains.
The proposal, published in the December issue of Nature Biotechnology, could also help biosimulation gain broader acceptance in the research community by ensuring that results of in silico experiments can be validated in the same manner as wet-lab methods.
"People publish the results of a model, describe the mathematical aspects, but often do not even bother to describe the tools they used, and they do not provide the model with the paper, so the referees or the readers cannot really assess the model."
"We very much hope that MIRIAM will foster a new evolution of the field, because a big problem in the field of quantitative modeling is that often the results are not reproducible," said Nicolas Le Novère, a researcher at the European Bioinformatics Institute and lead author on the paper. "People publish the results of a model, describe the mathematical aspects, but often do not even bother to describe the tools they used, and they do not provide the model with the paper, so the referees or the readers cannot really assess the model."
Despite MIRIAM's feminine moniker, Le Novère described the standard as "the son of SBML [Systems Biology Markup Language], because one side effect of SBML … was the formation of a community of developers, of modelers." SBML began in 2000 as a collaboration between seven research projects, Le Novère said. According to the project website (http://sbml.org/index.psp), more than 90 software packages now support the standard.
Le Novère, project lead for the BioModels database at the EBI, said that the motivation for MIRIAM grew out of a desire to "enhance and control the semantics of the models beyond the syntax" of SBML. While SBML enables researchers to exchange models, "even if you can exchange a model, and read a model, that doesn't mean that you can make sense of it," he said.
With biochemical models that are MIRIAM compliant, however, "If I read that a model describes the reaction between A and B, I can actually understand what A and B are in the mind of the person who developed the model," he said.
MIRIAM contains two parts: a set of guidelines that ensures that a model matches its description in a "reference correspondence," which is generally publication in a scientific journal. This part covers the "syntax and semantics" of the model, according to the Nature Biotech paper. The second component is a set of "annotation schemes" that includes information about the model's author, links to external bioinformatics resources that annotate components of the model, and other "external knowledge." (see boxes, below, for a more complete description of MIRIAM's two components.)
The aim of MIRIAM is to ensure that published research results derived from biochemical simulations are just as reproducible as those resulting from experimental platforms.
"We have to be able to run the model, to simulate it, and get the same results as the ones that were published in the description of the model," Le Novère said. "That seems kind of obvious when you describe it, but it's not at all the case, because sometimes you can, for instance, get the same result but with a different structure, and that's not right. And sometimes you reproduce exactly the model as it is described, and you discover that it does not behave the same way."
Currently, Le Novère said, MIRIAM covers "the so-called kinetic model of biochemistry and cell biology," which includes those that can be described by differential equations. He noted that there was a bit of "debate" among the 16 authors of the paper regarding the scope of the standard, however. "There is always a temptation to be comprehensive, but the more comprehensive you are, the harder the standard is to enforce," he said.
The proposal does not address stoichiometric analysis and other modeling methods. While these techniques may be included in future versions, Le Novère said that MIRIAM currently covers the most common class of biochemical models, "and those are the ones stored in the BioModels database or the CellML repository or other similar databases."
To encourage modelers to comply with MIRIAM, Le Novère said that the SBML project is developing an extension "to specifically encode the required annotation," which would automate the annotation process for those software packages that support SBML. In addition, the MIRIAM authors are compiling a list of URIs for data resources that can be used to annotate the components of the models. Le Novère said that the authors are developing a set of web services to link these data sources to the models, and vice versa.
Le Novère said that the BioModels database "is already accepting only MIRIAM-compliant models," and that the other authors of the paper representing resources such as the CellML repository, Sigpath, EcoCyc, JWS Online, RegulonDB, and DOQCS all agreed to make MIRIAM compliance "mandatory" in "the near future."
There are currently 44 models in the public version of BioModels, Le Novère said, with around 75 in the process of curation. However, he said, "the problem is the legacy we have more than 400 models in the pipeline, so it's really a problem of workforce now."
In that sense, he said, quick adoption of MIRIAM by the modeling community will greatly benefit efforts such as the BioModels, "because it will decrease enormously the time spent on the curation of each model."
Academic resources are not the only ones that see promise in the standard. Andrew Finney, a co-author of the paper and a software developer at Physiomics, said, "If a curated resource has adopted MIRIAM, like BioModels, then that's a significantly more useful resource for companies versus those resources that don't comply."
Finney was unable to comment on Physiomics' plans for adopting MIRIAM, however.
Bernadette Toner ([email protected])
MIRIAM Compliance, Part I: Requirements for Reference Correspondence
MIRIAM Compliance, Part II:
Source: Le Novère N, et al. Minimum information requested in the annotation of biochemical models (MIRIAM). Nat Biotechnol. 2005 Dec;23(12):1509-1515.