Proposal: Different languages for model specification

[dweindl]

Motivation

There are a number of formats for specifying models in systems biology, each with their specific strengths and weaknesses. PEtab version 1.0.0 only allows Systems Biology Markup Language (SBML) models. While SBML is supported by a large number of tools, there are good reasons to use other formats. For example, rule-based model formats (e.g., BioNetGenLanguage) permit more abstract and compact specification of models based on rules, which are generalisations of reactions. Therefore, and based on user request (#436), we propose to lift PEtab’s restriction to SBML models and allow arbitrary model formats.

Proposed changes

• Changes to the PEtab YAML file:

  • Change sbml_files to models
  • models entries will be model IDs (following the existing conventions for PEtab IDs) mapping to:
  • location: path / URL to the model
  • language: model format Initial set of model format identifiers (to be extended as needed):
    • SBML: sbml
    • CellML: cellml
    • BNGL: bngl
    • PySB: pysb
  • An additional entry for mapping tables (see below) is added

  Example:

  Before:

  format_version: 1
  parameter_file: parameters.tsv
  problems:
  - condition_files:
  - conditions.tsv
  measurement_files:
  - measurements.tsv
  observable_files:
  - observables.tsv
  sbml_files:
  - model1.xml

  After:

  format_version: 2.0.0
  parameter_file: parameters.tsv
  problems:
  - condition_files:
  - conditions.tsv
  measurement_files:
  - measurements.tsv
  observable_files:
  - observables.tsv
  mapping_file: mappings.tsv # optional 
  models:
    id_for_model1:
      location: model1.xml
      language: sbml

• Changes to the format of existing tables/files:

  • Condition/Observable/Parameter Table All symbols that previously referenced the ID of SBML entities, such as parameter IDs or compartment IDs, now refer to (globally unique) named entities in the model, such as parameters, observables, expressions. For example, condition table columns may correspond to parameters, states, species of the referenced model. For species, assignments in the condition table set the initial value at the beginning of the simulation for that condition, potentially replacing the initialization from preequilibration. For all other entities, values are statically replaced at all time points. For entities that assign values to other entities, such as SBML AssignmentRules, the value of the target of that rule is statically replaced at all time points.

• Additional files

  • Mapping Table: Mapping PEtab entity IDs to entity IDs in the model. This optional file may be used to reference model entities in PEtab files where the ID in the model would not be a valid identifier in PEtab (e.g., due to containing blanks, dots, or other special characters). The tsv file has two mandatory columns: petabEntityId, modelEntityId. Additional columns are allowed. modelEntityIds must be unique identifiers in the model. The mapping table must not map modelEntityIds to petabEntityIds that are also defined in any other part of the PEtab problem. modelEntityId may not refer to other petabEntityIds, including those defined in the mapping table. petabEntityIds defined in the mapping table may be referenced in condition, measurement, parameter and observable tables, but cannot be referenced in the model itself. For example, in SBML, local parameters may be referenced as $reactionId.$localParameterId, which are not valid PEtab IDs as they contain a . character. Similarly, this table may be used to reference specific species in a BGNL model which may contain many unsupported characters such as ,, ( or .. However, please note that IDs must exactly match the species names in the BNGL generated network file and no pattern matching will be performed.

Implications

• Tools need to check the model format and provide an informative message if the given format cannot be handled
• Validators will skip model-dependent validation when encountering unknown model types - ideally there would be some plugin mechanisms to provide validation

---

Co-authored by @FFroehlich @fbergmann. Also thanks to everybody participating in these discussions during the last COMBINE meeting.

[paulflang]

no pattern matching will be performed

So just that I understand this correctly, if I want to do pattern matching, I can still create a BNGL observable in the BNGL model like so

begin molecule types
    A(b~c~d)
end molecule types
begin observables
    Species myObservable A()
end observables
...

and use its ID myObservable in PEtab 2.0.0, for instance in the observable table (even without the mapping table, unless for whatever reason, I've decided to use an ID for the BNGL observable that contains characters that are not allowed in PEtab), but not in the condition table (as I would not know how that makes sense, also not if I map it to another petabEntityId in the mapping table, of course).

Anyway, awesome that PEtab will soon allow other model specifications. Thanks and congratulations!

[FFroehlich]

no pattern matching will be performed

So just that I understand this correctly, if I want to do pattern matching, I can still create a BNGL observable in the BNGL model like so

begin molecule types
    A(b~c~d)
end molecule types
begin observables
    Species myObservable A()
end observables
...

and use its ID myObservable in PEtab 2.0.0, for instance in the observable table (even without the mapping table, unless for whatever reason, I've decided to use an ID for the BNGL observable that contains characters that are not allowed in PEtab), but not in the condition table (as I would not know how that makes sense, also not if I map it to another petabEntityId in the mapping table, of course).

Yes, that is correct. In principle, you could also assign the value of the observable in the condition table, but there probably are only a few circumstances where this would make sense.

Anyway, awesome that PEtab will soon allow other model specifications. Thanks and congratulations!

Thank you!