The Common Fund Data Ecosystem’s Crosscut Metadata Model (CFDE C2M2) Roadmap

This document introduces the Crosscut Metadata Model (C2M2)], a flexible standard for describing biomedical experimental data. The Common Fund Data Ecosystem (CFDE) group is creating a new infrastructure, with C2M2 as its central concept, through which powerful cross-dataset searches, custom aggregation of experimental data and scale-powered statistical analysis methods will be made possible for the biomedical research community at an unprecedented scope.

Using this new infrastructure, the data coordinating centers (DCCs) can share structured information (metadata) about their experimental resources with the research community, dramatically widening access to usable observational data and accelerating discovery.

DCC Metadata Submissions

The DCCs will collect and provide metadata to CFDE describing experimental resources within their purview. Each metadata submission will take the form of a collection of tab-separated value (TSV) files. Precise formatting requirements for these TSV collections will be specified by JSON schema documents implementing the Data Package meta-specification published by the Frictionless Data group. These schemas will be used by the CFDE software infrastructure to automatically validate submission format compliance and metadata integrity during the metadata ingestion process.

The CFDE will offer the DCCs multiple alternatives for metadata submission formats, all of which will be automatically interoperable with the C2M2 ecosystem. These alternative formats are arranged in levels, tiered according to increasing complexity and reflecting anticipated differences in the relative richness of metadata available to different DCCs at any time. The general expectation will be that the metadata submitted and managed by a DCC will be able to transition, over time, through increasingly rich modeling levels as the life cycle of DCC/CFDE technical interaction progresses, which will enable increasingly powerful downstream applications.

C2M2 Richness Levels

In its current form, C2M2 is an entity-relationship system that models common properties of resources fundamental to biomedical research like subjects, digital files, events, biospecimens, and project datasets. Essential relationships between these fundamental resources are also formally described, documenting, for example:

  • the samples that were processed to produce a particular data file
  • which subject a given sample was drawn from (possibly obfuscated to protect patient privacy)
  • when a particular blood pressure measurement was made

While we know our current model is not yet rich enough to contain all the metadata that will be required by all DCCs, there are also too many terms for most DCCs to easily use. Modeling and data wrangling are always difficult, even for experts. Requiring every DCC to model their metadata using all possible features of the current C2M2 model as a precondition for submitting metadata to CFDE would be impractical for several important reasons.

  1. Many of the DCCs host human data, and for all but very basic file information, the associated metadata is protected. Currently, the CFDE does not have protected data access or a permission to make these metadata searchable. Requiring DCCs to supply these terms would make it illegal for most DCCs to comply.
  2. From an operational standpoint, the C2M2 model must remain as flexible as possible, especially during its developmental phases, to accommodate mutual learning between the DCCs and CFDE as the process of data ingestion develops. We do not want to lock ourselves in to our current model topology before we begin adding real data. It is far more expensive and error-prone to repeatedly change a complex model than it is to build one gradually from a simpler core concept that can stabilize before more specialized branches are added.
  3. Even if the issues with protected metadata were solved and the model was perfected, the complexity of the overall model would create avoidable and unnecessary onboarding delays for any new DCC.

With the design of C2M2, we are splitting the difference between the ease of evolution inherent in a simple model and the operational power provided to downstream applications by more complicated and difficult-to-maintain frameworks. DCCs with advanced, operationalized metadata modeling systems of their own should not encounter arbitrary barriers to CFDE support for more extensive relational modeling of their metadata if they want it; CFDE will maintain such support by iteratively refining the current C2M2 model according to needs identified while working with more operationally advanced DCCs.

Newer or smaller DCCs, by contrast, may not currently have enough readily-available information to describe their experimental resources using the most complex C2M2 modeling level. The CFDE will support cases like these by offering simpler but still well-structured metadata levels, lowering some of the barriers to rapid entry into the data ecosystem. We expect this concept of levels to be useful even after all current DCCs are onboarded: when the Common Fund funds new programs, they will all have ramp up phases where their data is less rich than the more mature DCCs.

Simpler C2M2 metadata levels must be maintained by the CFDE to maximize interoperability with more complex C2M2 variants; the whole system should be structured to minimize the negative side effects of overall model changes. These considerations have led to the creation of C2M2 richness levels: concentric, canonical subsets of C2M2 that are benchmarked at increasing levels of model complexity and detail, wherein each successive modeling level is a value-added superset of all of the metadata encompassed by the previous (less complex) level.

Presently, the CFDE offers two less complex C2M2 variants in addition to the most complex, current C2M2 model: Level 0 (basic metadata describing a collection of digital files) and Level 1. Level 1 accomplishes the following:

  • Introduces terms for core experimental resources, like samples and subjects.
  • Provides a rudimentary set of search targets in the form of annotations, like the anatomical location of the source for a human tissue sample or taxonomic data describing sample source organisms and study subjects.
  • Supports arranging experimental resources into sub-collections based on a hierarchy of projects, studies, or other similar subdivisions of research ownership and responsibility.

Levels 0 and 1 consist entirely of unprotected metadata terms, and therefore should be achievable by all DCCs. These two levels are sufficient to support two minimal Use Cases. Using only the metadata from Levels 1 and 2, a researcher can find datasets from across the Common Fund that contain information about her tissue of interest, assayed using her method of choice. Similarly, staff at the NIH can compare the tissues, species and assay types available across DCCs. More complex Use Cases, such as the ability for a researcher to find datasets for patients with a specific disease in a specific age range will require the DCCs to submit Level 2, protected, metadata terms. However, this is limited by our access to protected data, as we cannot legally accept that data until certain security and administrative issues are solved.

C2M2 Development Roadmap

Level 0

C2M2 Level 0 defines a minimal valid C2M2 instance. Data submissions at this level of metadata richness will be the easiest to produce and will support the simplest available functionality implemented by downstream applications.

The full specification can be found in the accompanying Level 0 popout document.

Level 1

Level 1 introduces tables for core experimental resources like:

  • Samples and subjects
  • Search targets in the form of annotations, like the anatomical source for a given tissue sample
  • Host species taxonomy for samples and subjects
  • Basic support for arranging experimental resources into sub-collections based on a hierarchy of projects or studies

The full specification can be found in the accompanying Level 1 popout document.

(Level 2)

Level 2 introduces tables for core experimental resources like:

  • Sex
  • Age
  • Disease
  • Health conditions
  • Vital stats. e.g. height, weight, BP, etc.

We have a draft specification of Level 2; preliminary documents for the schema can be found here. Finalization of Level 2 will be completed to achieve a CFDE portal demonstration in December 2020. Level 2 can be used to represent protected data, which will require the completion of several important administrative and policy milestones.

(Level 3)

Level 3 introduces tables for core experimental resources like:

  • Sequencing technology, e.g. Illumina, nanopore, 454
  • Geographic location
  • Race/ethnicity (human)/strain (mouse)
  • Sample collection date
  • SOP used for extraction or analysis process

The full specification is currently outlined, but not under active development. It will be formalized in the coming months and implemented in a later demo. As with Level 2, it is dependent on CFDE access to protected data.

(Level N)

Based on feedback from the DCCs, and the terms that are important for searching their data, we anticipate the need for further levels. Until we begin receiving metadata from the programs, we cannot predict what terms will be added or what constellations of terms will be required to make a compliant model. However, we do not expect the number of levels to exceed 5, as some new terms will best fit as amendments to previously defined levels.

The full specification is expected, but not outlined or under active development. It will be outlined once the DCCs begin to submit data and formalized in collaboration with our DCC partners. As with Level 2, implementation of these levels will be dependent on CFDE access to protected data.