The ISOP technology committee recently requested feedback through a survey with the topic “How do you share?” (http://discuss.go-isop.org/t/isop-global-survey-how-do-you-share/657) asking how people share models and modeling scripts. At the time of writing, over 100 people have filled in responses. There is a split of about 50 / 50 between academic and industrial respondents.
Of particular interest to DDMoRe are the questions about repositories used in sharing models and scripts. Fortunately, the majority of respondents have heard about the DDMoRe model repository (http://ddmore.eu/product/model-repository). However, very few respondents had actually used our repository for model or code sharing yet, with Github cited as the principle means of sharing code. Of some comfort is the response that looking forward within the next two years, the DDMoRe repository “catches up” with Github as a means for model or code sharing.
As a Github user, I can appreciate that it provides good tools and means for sharing and for collaborative working on code, but I have never considered sharing models via that platform. While models are routinely expressed as code (NMTRAN, MLXTRAN, R scripts), when it comes to sharing models it takes a considerable amount of additional information to convey the meaning, scope, context of use, assumptions about a given model. After all, whole manuscripts are written to describe models.
It is in providing this context, annotation, curation and additional metadata that the DDMoRe repository starts to pull ahead of other tools like Github, Google Drive, etc. Our repository is designed to provide much of this background information and enable the modeller browsing the repository to assess whether a model is appropriate for their use, perhaps as a starting point for further development. Like the Biomodels database for systems biology and quantitative systems pharmacology, the DDMoRe repository is designed to facilitate browsing and discovery of models for particular disease areas and endpoints. The curation of models within our database adds additional assurance that the model meets the necessary criteria for quality and reproducibility (http://www.ddmore.eu/product/model-certification). The DDMoRe repository is open to anyone to browse, download and use models and also to upload models either using DDMoRe standards (PharmML and MDL) or with original code e.g. NMTRAN, MLXTRAN. Models within the repository must be executable and the modeller uploading to the repository must provide files to allow the model to be run and results checked against reference output.
Sharing reproducible models and code is key to moving science forward and preventing re-invention of the wheel. It is encouraging that ISOP is finding acceptance from the survey respondents in sharing. Naturally we want modellers to consider the DDMoRe repository as their key repository in sharing models and code for reproducibility of results, and we look forward to more models being submitted and curated within our repository.