Scientific collaborations involving multiple institutions are increasingly commonplace. It is not unusual for publications to have dozens or hundreds of authors, in some cases even a few thousands. Gathering the information for such papers may be very time consuming, since the author list must include authors who made different kinds of contributions and whose affiliations are hard to track. Similarly, when datasets are contributed by multiple institutions, the collection and processing details may also be hard to assemble due to the many individuals involved. We present our work to date on automatically generating author lists and other portions of scientific papers for multiinstitutional collaborations based on the metadata created to represent the people, data, and activities involved. Our initial focus is ENIGMA, a large international collaboration for neuroimaging genetics.
Significant scientific effort is devoted to describing data with appropriate semantic metadata. Many communities have data repositories that use semantic markup to describe datasets, enabling users to retrieve data based on metadata properties of interest. In neuroimaging, which is the focus of this work, neuroinformatics repositories exist (e.g., http://nitrc.org) where researchers may download brain scans corresponding to individuals of a certain age range. However, this metadata has been used in very limited ways beyond the repositories. Once the datasets are extracted from a repository, they often become separated from their metadata when they are analyzed in a separate system. Published articles include citations to the datasets that contain a unique identifier provided by the original repository, however their original semantic metadata is usually not passed on and is only informally described in the articles.
We are interested in the use of semantic metadata to automatically generate portions of scientific papers that describe datasets included in the publication. For example, biomedical papers that use datasets collected from a group of study participants often include demographic tables with information that often exists in the metadata for the datasets (e.g., age ranges, clinical characteristics). Similarly, metadata for all the datasets used may include pointers to the investigators who collected and contributed the data, who would often be included as authors of a paper using the dataset. In large multi-institutional collaborations where papers include dozens or hundreds of authors, generating the author list by hand can be very tedious.
This paper presents our approach to generating portions of scientific papers based
on a semantic repository of project information. We created a semantic repository of
projects, contributors and datasets, and used it to automatically generate author lists
and descriptions of datasets. We use the Organic Data Science framework that we
developed in prior work [
In other work, we developed an approach to automatically generate the methods
sections of papers from scientific workflows and their associated metadata [
This paper starts with a description of what portions of the papers could be generated automatically from a semantic repository of project information. We also describe the ontology and semantic repository that we developed to represent the information for ENIGMA. We then present our approach to generate portions of scientific papers, and show a detailed example of a representative ENIGMA paper. 2
To illustrate the potential uses of semantic metadata to automatically generate portions of papers, we use several examples of publications by the ENIGMA Consortium, which involve many international research groups.
Author lists are often organized based on the roles and contributions made to the
work. Consider [
Gathering information about authors and study metadata for papers from
multiinstitutional collaborations is a very tedious process. Authors come from numerous
institutions around the world, and each may have multiple institutional and
department affiliations. Almost 300 authors and 200 institutions are listed in [
The manuscripts themselves include tables with information about the datasets
used. In ENIGMA, papers often report clinical associations with medical (brain)
image features, pooled from dozens of individual imaging studies around the world. It is
therefore typical to include tables of the data collected in each study (a cohort), or the
details of the data image collection process (an acquisition protocol). These tables
include the information for the brain scanner used, demographics of the cohorts
involved and the inclusion and exclusion criteria for data in each study cohort. Other
tables may include other data summaries, such as genotyping platforms, or diagnostic
scales that may be more specific to a particular project. These tables provide
important provenance information, and may not be identical across clinical focus areas.
Table 1 shows an excerpt of a table of acquisition protocols for [
The examples given here are representative data-rich aspects of manuscripts that
may be automatically generated. Although we focus on ENIGMA, large scale
collaborations are becoming a key aspect of data discovery in the biomedical and broader
scientific research fields. Our requirements are shared by large multi-institutional
scientific collaborations, such as the climate collaboration described in [
This section describes the ontology that we created to describe large multiinstitutional collaborations and its use in a semantic repository for ENIGMA. 3.1
The collaboration ontology for the ENIGMA consortium was created to represent information that is crucial to organize the different activities, datasets and contributors. The classes and properties were created to represent information that is important hasMember
usesDataset to show in the manuscripts that result from collaborative activities. We created an initial collaboration ontology to fit the needs of ENIGMA
Figure 1 shows the main concepts of the ontology, which include Working Group, Project, Dataset (collected for a group of people or cohort), Acquisition Procedure, and Person. It also shows the relationships between them. While some ontologies focus on people and projects and others focus on describing datasets, this ontology combines both aspects of the collaboration. We extended this core ontology with classes and properties that are used to describe projects, datasets, and people in ENIGMA. The ontology could be extended similarly for other collaborations by creating properties of datasets and acquisition procedures in their particular domain.
To generate the author list, the contributors of the project, along with their role for that particular project, must be known. Thus, the “hasPSeniorLead,” “hasPJuniorLead,” and “hasPSupportingContributor” properties of the Project class allow the system to find out who needs to be acknowledged in the author list. In addition to the leadership roles of the project, the people who took charge of the cohorts are also acknowledged. Therefore, the Cohort class has two properties that represents who the principal investigator and the other investigators: “hasPI” and “hasInvestigator.”
A table with image acquisition protocols is necessary to include in all manuscripts that include imaging data, to describe the data collection procedure for each cohort. The image acquisition protocol table currently varies by paper but the most common columns were made into properties in the ontology. The “hasAcquisitionProcedure” property relates an image acquisition protocol to its respective cohort. The “AcquisitionProcedure” class has a “ImageAcquisitionProtocol” subclass with the following properties, which are used to generate a table summarizing the image acquisition protocols: “hasAcquisitionDirection,” “hasSequence,” “hasScanner,” “hasDataAcquisitionMatrix,” “hasFlipAngle,” “hasFoV,” “hasNumberOfEchoes,” “hasNumberOfSlices,” “hasScanTime,” “hasSliceThickness,” and “hasVoxelSize.”
A table of participant demographics is common for all manuscripts that study human populations, and therefore a key component of ENIGMA papers. Datasets included in ENIGMA are collected from a cohort, i.e., a group of people who participate in a study; it is therefore important to consistently describe cohort demographics for each dataset included in a study (manuscript). To retrieve the total number of participants in a cohort, the number of males, and the number of females of a cohort, one can use the “hasNumberOfParticipants,” “hasNumberOfMales,” and “hasNumberOfFemales,” respectively. Since it is often necessary to describe separately the group of patients and the group of controls in a study cohort, a new class called “CohortGroup” was created. Thus, a Cohort has the properties “hasControlGroup” and “hasDiagnosticGroup” with “CohortGroup” as their range. The “CohortGroup” has a “hasNumberOfParticipants,” “hasNumberOfMales,” and “hasNumberOfFemales” as properties, which are necessary to generate the cohort demographics table.
Inclusion and exclusion criteria is also typically reported in a paper that describes a recruited set of participants. Therefore, it is important for ENIGMA papers to include a table with this information for all cohorts, as it is a key aspect of many clinical studies of a particular patient population. The inclusion and exclusion criteria are described through a series of properties for the Cohort class. The criteria are not properties of the Cohort class as the control group and diagnostic group of a cohort can have different inclusion and exclusion criteria. Thus, separating the inclusion and exclusion criteria for the patient group and the control group was necessary. The “CohortGroup” properties that begin with “has” are inclusion properties while the “CohortGroup” properties that begin with “didNot” or “doesNot” are exclusion properties. The following properties are example properties that describe the inclusion criteria for a specific cohort group: “hasDisorder,” “hasFirstDegreeRelativeWithDisorder,” “hasFirstEpisodeOf,” “hasMedicalRatingDetails,” “usesTreatment,” “hasTreatmentDetails,” “hasNeurologicalComorbidity,” “isClinicallyStable,” and “isProficientinLocalLanguage.” The following example properties describe exclusion criteria: “didNotHaveFirstEpisodeOf,” “doesNotHaveDisorder,” “doesNotUseTreatment,” “doesNotHaveContraindicationToMRI,” “doesNotHaveNeurologicalComorbidity,” “isNotPregnant,” and “doesNotHaveIntellectualDisability.”
Many of these properties have an inverse. For example, the property “hasSeniorLead” has domain project and range person, and “isSeniorLeadOfP” is its inverse. There are also additional properties that describe the ENIGMA concepts further. For example, a project has a “hasApprovedProposalForm” property that links to the proposal that the project leads originally submitted to describe the project and have it approved.
3.1
We use the Organic Data Science framework [
The ENIGMA repository is being prototyped with 3 selected projects out of more than 50 ENIGMA projects, with a total of 405 pages. It includes 3 working groups, 3 projects, 4 image acquisition protocol pages, 8 scanners, 89 cohort groups, 54 cohorts, and 112 persons. We continue to grow the repository so that all of the working groups, projects, cohorts, and researchers will be eventually represented. The repository is currently private to ENIGMA members 4
There are different approaches to organizing an author list. In ENIGMA, the following are two typical approaches. One approach is based on fine grained contribution, where the authors are ordered based on their roles. Under each role, the authors are alphabetically ordered. Another approach is based on coarse grained contribution, where the authors from roles other than junior and senior leads are placed in the middle of the list in alphabetical order. The junior and senior leads do not follow alphabetical ordering in either approach.
To generate the list of authors automatically, the system extracts the names and details of researchers involved in a project by leveraging the ENIGMA ontology. The list of authors is generated using one of the two approaches described above.
In addition to the author list, our system generates a detailed credit section that is often included in the acknowledgements of papers listing each individual’s contribution to the work. Authors who have multiple roles are credited for all of their roles. Cohort Data Type Scanner
T1-weighted 3T Magnetom AB MRI TIM Trio
T1-weighted 1.5T MagneXY MRI tom Sonata
The system automatically generates three types of tables: an imaging acquisition protocol table, a demographics table, and an inclusion/exclusion criteria table.
Table 2 shows an example of an automatically generated image acquisition protocol table. Image acquisition protocol tables display the information regarding the MRI scanner and imaging sequence used to scan each cohort. Here, the CLING acquisition protocol metadata from Figure 2 was used.
Table 3 shows an example of an automatically generated demographics table. Demographics tables show the general information about the study participants in each cohort. Here, our table displays the cohort, diagnostic information (if applicable), the total number of individuals in each cohort broken down to the number of males and females, and the average age and age range of each group.
Inclusion/exclusion criteria tables include, for each cohort, information on any inclusion or exclusion criteria used for enrolment in a study. In MRI studies, individuals are excluded for having metal implants which may interfere and cause harm when placed inside a high magnetic field (i.e., MRI machine). Due to space constraints, we have omitted an example of this type of table. Voxel Size 1 mm^3 1 mm^3 5
The author generation system assumes that separate pages are maintained for each author with details on their full name, their full set of affiliations, their highest degree, and their contact email (used only for a corresponding author). One limitation of the current application is that it assumes the authors are individuals and does not address cases where a consortium or group are included as authors. It also assumes that the author list always follows one of the two author list approaches described above, yet there can be many other possible approaches to author ordering. In addition, author information could be obtained from existing repositories such as ORCID or VIVO.
Currently our system can only generate three types of tables containing certain preselected columns. We envision custom table formats being created by project leads, so that they can easily be used by other projects within the ENIGMA consortium.
Significant amounts of information about ENIGMA are available in unstructured form. For example, the image acquisition protocols and the inclusion/exclusion criteria are not structured. The ENIGMA ontology allows for the metadata to be harmonized and reported in a standardized way. Now the ENIGMA repository has the potential to go beyond table generation for papers and allow filtering and selecting datasets by using standardized metadata. We believe these efforts will allow for improved collaborations and scientific discovery. 6
As scientific collaborations become more complex, documenting the details of the datasets used becomes increasingly challenging since it involves gathering information from dozens or hundreds of individuals across many institutions. We have shown that a semantic metadata repository for a collaboration enables the creation of tools to generate automatically author lists and tables that summarize key information about data collection and other data characteristics that are important to an article. We have an initial implementation of a semantic repository and associated generation tools for the ENIGMA neuroimaging genetics collaboration, which we continue to extend both in content and capabilities.
Acknowledgements. We are very grateful to the KAVLI foundation for their support of ENIGMA Informatics (PIs: Jahanshad and Gil). We also acknowledge support from the National Science Foundation under awards IIS-1344272 (PI: Gil), ICER1541029 (Co-PI: Gil), and IIS-1344272 (PI: Gil), and from the National Institutes of Health’s Big Data to Knowledge Grant U54EB020403 for support for ENIGMA (PI: Thompson). We thank the members of the Organic Data Science and Linked Earth projects for their contributions to the design of the framework. We also thank the many participants of the ENIGMA collaboration for their feedback to this work.