Fuzzy locations break document validity

[jakebeal]

The current implementation of the SBOL/Genbank converter in sbol3_genbank_conversion.py uses a custom class Location_GenBank_Extension to represent GenBank's fuzzy location field.

Unfortunately, the current implementation breaks document validity, because it tags the objects with type http://sbols.org/v3#locationPosition, which is not a valid SBOL type. This causes any SBOL library not already adjusted to accept the extension (e.g., pySBOL3) to deem the document invalid (and possibly also crash during loading, depending on implementation).

Changing the intended type for the location, http://www.ncbi.nlm.nih.gov/genbank#locationPosition, does not fix the problem either, because now there is a SequenceFeature whose hasLocation property points to something that the library doesn't believe is an SBOL Location.

[jakebeal]

A workaround for now will be to simply not use this class: instead, use the same behavior of the old converter, which just truncated fuzzy ranges into sbol:Range objects.

Some possible better resolutions for the future:

• Get fuzzy locations accepted into the SBOL specification (feature request: Range location object to have support to store fuzzy endpoints SBOL-specification#491) - though this will still have backward compatibility problems
• Put a range in parallel with a fuzzy range linked via a NCBI:hasFuzzyLocation property (so that it doesn't trigger validation)

[jakebeal]

On deeper inspection, it looks like we're already partway to solution via the section approach, since the offending SequenceFeature objects are linked via NCBI:fuzzyFeatures. So if the SequenceFeature was changed to an extension class NCBI:fuzzySequenceFeature, that would be sufficient to interrupt the invalidity.

[tcmitchell]

Is this possible a job for an extension? Could this work by extending Location or one of its children using the pySBOL3 CustomIdentified class? See https://pysbol3.readthedocs.io/en/stable/extensions.html#example-2-extend-a-core-class for more info.

[jakebeal]

Unfortunately, I don't currently see how to do it:

• The fuzzy range semantics are not compatible with any of the children. Range is closest, but will fail because there are validation rules based on the length of a range, which requires exact start and end. This is why we didn't extend Range in the first place.
• Because Location is an abstract class, it's not clear whether any library that isn't using the extension can meaningfully instantiate a generic Location. In pySBOL3, for example, there is no builder registered for Location, so extending Location results in errors when working with a document with pySBOL3 (i.e., without the extension loaded).

[tcmitchell]

That's unfortunate. I'm glad you explored the possibility. I have opened a pySBOL3 issue to understand this limitation better: SynBioDex/pySBOL3#427

[jakebeal]

Patch has been merged; issue will stay open for a better resolution.