sep_019.md

SEP 019 -- Using SBOL to model the Design-Build-Test-Learn Cycle

SEP
Title	SEP 019 -- Using SBOL to model the Design-Build-Test-Learn Cycle
Authors	Bryan Bartley (bartleyba@sbolstandard.org), Jake Beal (jakebeal@gmail.com), Raik Gruenberg (raik.gruenberg@gmail.com), James McLaughlin (j.a.mclaughlin@newcastle.ac.uk), Chris Myers (myers@ece.utah.edu), Nicholas Roehner (nicholasroehner@gmail.com), and Anil Wipat (anil.wipat@newcastle.ac.uk)
Editor	Nicholas Roehner (nicholasroehner@gmail.com)
Type	Data Model
SBOL Version	2.3
Replaces	SEP 014, SEP 016, SEP 017, SEP 018
Status	Accepted
Created	10-Nov-2017
Last modified	11-Dec-2017
Issue	#43

Abstract

Linking experimental data with SBOL designs is becoming critical to a number of important synthetic biology projects. Therefore this SEP introduces a data model for SBOL that supports Design-Build-Test-Learn reasoning. As the synthetic biologist proceeds around the Design-Build-Test-Learn cycle, each iteration generates new understanding that must be integrated at each stage. This SEP will help synthetic biologists close the loop in the Design-Build-Test-Learn cycle.

Table of Contents

• 1. Rationale
• 2. Specification
  • 2.1 Implementation
    • 2.1.1 Implementation.built
  • 2.2 Design, Build, Test, and Learn
  • 2.3 Best Practices
    • 2.3.1 Usage Roles
    • 2.3.2 Versioning versus Provenance Semantics
  • 2.4 Validation Rules
• 3. Examples
  • 3.1 Use Cases
  • 3.2 Design-Build-Test-Learn Provenance
  • 3.3 Assembly Provenance
  • 3.4 Implementation
  • 3.5 Attachment
• 4. Backwards Compatibility
• 5. Discussion
• 6. Relation to Other SEPs
• References
• Copyright

Rationale

• Decouple the design-build-test-learn process, according to the foundational principles for engineering biology.
• Specify where to add experimental data
• Provide an SBOL representation of biological instances of a design that can link to LIMS systems
• Provide clear guidance for using PROV-O with illustrative examples.
• Capture workflow provenance, a description of the events of a workflow, which is crucial for scientific reproducibility
• Support model-based design.

This SEP was initiated in response to the ["Design-Build-Test" thread] on sbol-dev.

Specification

Here we propose to add a new class, Implementation, to allow users to represent physical realizations of biological designs. For example, an Implementation could be used to represent an aliquot of DNA, a cell clone, or a lysate. The Implementation class is meant to serve as a connection point between theoretical designs of biological systems and descriptions of their actual structure and/or function following their physical construction.

Here we also define four new SBOL ontology terms (sbol:design, sbol:build, sbol:test, and sbol:learn) to serve as values for the hadRole properties of the PROV-O classes Usage and Association. In addition, we present tentative best-practice validation rules for the use of these ontology terms and PROV-O to represent design-build-test-learn cycles. These best practices will undergo continuing development to support new use cases as they arise.

2.1 Implementation

An Implementation represents a real, physical instance of a synthetic biological construct which may be associated with a laboratory sample. An Implementation describes the thing which was built during the Build phase of a D-B-T-L workflow. An Implementation may be linked back to its original design (either a ModuleDefinition or ComponentDefinition) using the wasDerivedFrom and/or wasGeneratedBy properties inherited from the Identified superclass. An Implementation may also link to a ModuleDefinition or ComponentDefinition that specifies its actual realized structure and/or function as described in Section2.1.1.

Figure 1: Diagram of the Implementation class and its associated properties

2.1.1 Implementation.built

The built property is OPTIONAL and MAY contain a URI that MUST refer to a TopLevel object that is either a ComponentDefinition or ModuleDefinition. This ComponentDefinition or ModuleDefinition is intended to describe the actual physical structure and/or functional behavior of the Implementation. When the built property refers to a ComponentDefinition that is also linked to the Implementation via PROV-O properties such as wasDerivedFrom, it can be inferred that the actual structure and/or function of the Implementation matches its original design. When the built property refers to a different ComponentDefinition, it can be inferred that the Implementation has deviated from the original design. For example, the latter could be used to document when the DNA sequencing results for an assembled construct do not match the original target sequence.

2.2 Design, Build, Test, and Learn

The ontology terms design, build, test, and learn are OPTIONAL values of the hadRole properties of the Usage and Association classes. These properties describe how entities (such as samples, data, or models) are used in an Activity and what role an Agent or Plan (such as a person, software tool, or protocol) plays in an Activity, respectively.

In natural language, these terms indicate the following:

• "design" describes a process by which a conceptual representation of an engineer's imagined and intended design is derived, possibly from a model or pre-existing design.
• "build" describes a process by which a sample is derived in accordance with a conceptual design, often from other samples.
• "test" describes a process by which raw data or observations are derived via experimental measurement of samples.
• "learn" describes a process by which a theoretical model, analysis, datasheet, etc. is derived, usually from experimental data or another model.

2.3 Best Practices

2.3.1 Usage Roles

To facilitate data exchange across different domains of synthetic biology, a user MAY use one of the terms "design", "build", "test", or "learn" to specify Usage roles.

A user MAY also specify additional Usage roles that correspond to their own home-made ontologies for specifying recipes and protocols.

2.3.2 Versioning versus Provenance Semantics

A new object which is the product of an Activity links back to the Activity which generated it via the wasGeneratedBy field. By the W3 PROV-O specification, generation is defined as:

...the completion of production of a new entity by an activity. This entity did not exist before generation and becomes available for usage after this generation.

Provenance semantics are somewhat different from the versioning semantics defined in the SBOL specification. The SBOL specification defines a new version of an object as an update of a previously published object (and therefore a previously existing object). In contrast, an SBOL object which is "generated" from another SHOULD BE regarded as a new entity, not a new version by the PROV-O specification above. However, this distinction is somewhat subjective (see Theseus paradox). Therefore we RECOMMEND as a best practice that objects linked by Activities not be successive versions of each other, though we leave this at the discretion of users and library developers.

2.4 Tentative Validation Rules

A design-build-test-learn process would typically generate new SBOL objects in the order of one or more ModuleDefinitions and/or ComponentDefinitions, followed by one or more Implementations, followed by one or more Collections of Attachments, followed by one or more Models. This order of operations (among others) is compatible with the following best-practice validation rules:

• An Activity that contains an Association of role "design" SHOULD be referred to by the wasGeneratedBy property of at least one ModuleDefinition, ComponentDefinition, or Sequence. If this Activity contains one or more Usage objects, then at least one of them SHOULD be of role "learn" or "design".

• An Activitythat contains an Association of role "build" SHOULD be referred to by the wasGeneratedBy property of at least one Implementation. If this Activity contains one or more Usage objects, then at least one of them SHOULD be of role "design" or "build".

• An Activitythat contains an Association of role "test" SHOULD be referred to by the wasGeneratedBy property of at least one Collection of Attachment objects. If this Activity contains one or more Usage objects, then at least one of them SHOULD be of role "build".

• An Activity that contains an Association of role "learn" SHOULD be referred to by the wasGeneratedBy property of at least one Model, Collection of Attachment objects, ModuleDefinition, ComponentDefinition, or Sequence. If this Activity contains one or more Usage objects, then at least one of them SHOULD be of role "test".

• A Usage of role "design" SHOULD refer to a ModuleDefinition, ComponentDefinition, or Sequence.

• A Usage of role "build" SHOULD refer to an Implementation.

• A Usage of role "test" SHOULD refer to a Collection of Attachment objects.

• A Usage of role "learn" SHOULD refer to a Model.

Examples

See Discussion for an in-depth explanation of these examples.

3.1 Use Cases

Example 1: A hypothetical workflow for model-based design that demonstrates a variety of use cases which are supported by this proposal

3.2 Design-Build-Test-Learn Provenance

Example 2: A reproducible workflow for model-based design represented using PROV-O classes and annotations

3.3 Assembly Provenance

Example 3: A detailed representation of assembly provenance. Assembly provenance allows a synthetic biologist to track all the physical components or samples which go into a DNA assembly protocol.

3.4 Implementation

Example 4: An Implementation links a design ComponentDefinition to its build. In the left example, the build does not match its design specification. In the right hand example, the build does match its specification.

3.5 Derivation of Biological Replicates for iGEM Interlab Study

Example 5: An initial sample kit based on part designs from the iGEM Registry is replicated and split into different samples.

3.6 Co-Transfection of Constructs for CRISPR Repression Module

Example 6: Two sets of overlapping constructs that implement the components of a CRISPR-based circuit are transfected to implement two different versions of the circuit.

Backwards Compatibility

This proposal does not affect backwards compatibility.

Discussion

5.1 Design-build-test-learn

The design-build-test-learn cycle is a common theme in synthetic biology and engineering literature. The cycle is a generalized abstraction of an ideal engineering workflow and approach to problem-solving. Thefefore, the design-build-test-learn cycle is a de facto ontology upon which to base an SBOL data model for workflow abstraction. Other workflow activities in synthetic biology, such as analyzing, modeling, verifying, and evolving, by and large fit into the design-build-test-learn abstraction.

This specification enables synthetic biologists to capture workflow provenance with SBOL. The aim is to capture a complete description of evaluation and enactment of computational and laboratory protocols in a workflow. This is crucial to verification, reproducibility, and automation in synthetic biology. By specifying a coherent model for workflow provenance, we can also address other critical use cases such as specifying where to add experimental data and performing model-based design.

Provenance classes were previously adopted in SBOL 2.1, but a gap in the specification left some ambiguity in how these classes should be used. The 2.1 specification indicates that the hadRole field on Usage and Activity classes MUST be provided, but does not define any terms to use in these fields. Because such terms are not specified, it is currently not possible for tool-builders to interpret or exchange provenance histories. It is expected that users will use their own ontology terms to specify how objects were used in a recipe, protocol, or computational analysis. However, these home-made ontologies will be very domain specific and may not be intelligible to users working in another domain. For example a modeler should not be expected to understand an ontology of Usage roles for DNA assembly. The terms "design", "build", "test", and "learn" provide a high level workflow abstraction that allows tool-builders to track provenance within their domain as well as to track the flow of data between domains.

Example 1 illustrates the various use cases that are possible with this provenance representation. Use case 1 describes linking many biological instances (plasmid clones, cellular clones, etc.) to the design which generated them. Use case 2 describes linking sequential stages of a build process which requires more than one processing step in the laboratory. Use case 3 describes a simple case of linking experimental data, such as sequencing data, to a sample. Use case 4 describes performing an experimental test on multiple samples at once in batch, for example in a 96-well plate. Use case 5 describes a scenario in which different models (eg. deterministic vs stochastic) are derived from experimental data. Use case 6 describes a case in which data from multiple experiments are integrated into a single analysis. Use case 7 describes the creation of a model-based design.

The Design, Build, and Test objects illustrated in Example 1 are included for purposes of discussion, but this specification does not actually introduce new classes by these names. Rather, this specification introduces the Implementation class to represent "Builds". Also this specification introduces Attachments in order to wrap experimental data files generated by a "Test". A "Design" is represented by a ModuleDefinition.

Example 2 shows explicitly how SBOL objects can be linked via PROV-O in an idealized design-build-test-learn workflow. This particular example represents model-based design. The workflow begins with a Model which is used to generate a ModuleDefinition ("Design") using a computational tool such as iBioSim. This ModuleDefinition is then used to generate a new Implementation ("Build") via an AssemblyProtocol. Now that a "Build" has been constructed in the laboratory, a "Test" may be performed by running a MeasurementProtocol on laboratory equipment, thus generating a Collection of data files or Attachments. Finally, a new Model might be derived from these data. This Model may not match the beginning Model, as our observation may not match the prediction.

For illustrative purposes, Example 2A shows how Agent or Plan might be extended to represent the important players in a synthetic biology worflow. For example, a ComputationalDesignProtocol may document that a Python script was executed in order to optimize a DNA sequence for synthesis. An AssemblyProtocol may describe a laboratory protocol (for example, either manual or automated). A SoftwareTool or Person are Agents that assist in an Activity. These examples illustrate where a developer might want to hook in their own application-specific data model. The PySBOL and libSBOL libraries allow users to define their own extension classes to Plan and Agent through inheritance relationships.

In addition to specifying a procedural workflow, this specification also allows us to interpret other kinds of provenance. Example 3 illustrates an example of "assembly provenance". This allows the user to track all of the physical instances of components which may be used in a DNA assembly procedure as well as the original design used to generate the new build. Other kinds of provenance are also possible. For example, provenance might also be used to track all the Models used to compose a more complicated Model.

Example 4 illustrates how the Implementation class is used to differentiate a conceptual design from a realized design. In the left hand example, the Implementation links via PROV-O to the ComponentDefinition that represents the original design, while its componentDefinition property links to the ComponentDefinition which describes the structure of the actual physical implementation. The right hand example demonstrates a case where the Implementation matches its design specification.

Example 5 provides an example of representing a sample history in SBOL. An initial sample kit based on part designs from the iGEM Registry is replicated and split into separate kits for each iGEM team. Multiple samples are then extracted from each team's kit for further construction and/or testing. Software tooling can trace the provenance of each sample to determine whether its original design (the ComponentDefinition from which its parent sample was derived) matches the description of what was built (the ComponentDefinition linked via its built property).

Example 6 provides an example of representing physical composition with SBOL. Two sets of overlapping constructs that implement the components of a CRISPR-based circuit are transfected to implement two different versions of the circuit.Software tooling can confirm that a given circuit Implementation satisfies its original design in two ways. First, a tool can compare the ModuleDefinition objects referred to by the wasDerivedFrom and built properties of the circuit Implementation. Second, a tool can confirm that the ComponentDefinition objects composed by the ModuleDefinition are the same as those from which the circuit component Implementation objects are derived.

Relation to Other SEPs

This proposal evolved from SEP 14 after discussion at 2017 HARMONY and COMBINE workshops, on the Github issue tracker, and on the ["Design-Build-Test" thread]: https://groups.google.com/forum/#!topic/sbol-dev/AnpwJP2_f5A

This proposal is competing with SEP 20. The most significant difference between these SEPs is that SEP 20 includes a mandatory design field on its Implementation class. This field is meant to serve as the primary means for documenting the intended structure and/or function of an Implementation, rather than PROV-O classes.

On the one hand, the design field could serve as a simpler, more explicit means of expressing these semantics. On the other hand, the design field could be seen as redundant given SBOL's adoption of a subset of PROV-O. In addition, the design field is currently mandatory as specified by SEP 20, which fails to account for use cases where there is no intended design, such as a sample of natural/unknown origin or one produced via directed evolution.

If SEP 19 passes, it would not preclude the addition of an optional design field to the Implementation class in the future.

References

Copyright

To the extent possible under law, SBOL developers has waived all copyright and related or neighboring rights to SEP 007. This work is published from: United States.