I notice that after I run the reasoner, all projection neurons are listed as subclasses of Gigantocellular reticular nucleus projection neuron. It doesn't look like a location was defined for that cell type, which is probably the reason why.

Ah lack of test coverage. Serializing doid.owl multiple times revealed this.

Two olfactory cells have assertions that they are subclasses of gigantocellular reticular neurons:

'Olfactory bulb (main) granule cell' SubClassOf 'Gigantocellular reticular nucleus intrinsic neuron'

'Olfactory cortex large multipolar cell' SubClassOf 'Gigantocellular reticular nucleus intrinsic neuron'

Beyond ontload, ontrefactor, and necromancy, there are a number of oneliners and other functionality that should be consolidated into a single file, many from https://github.com/SciCrunch/NIF-Ontology/blob/uri-switch/docs/processes.md. Some scripts such as make_catalog could be subsumed.

One thing that is not in the docs at the moment that needs to be implemented is the ability to make a ttl/reasoner-subset.ttl from a glob/urls so that we reason about collections of ontologies that are far apart in the import chain and whose only common denominator is nif.ttl or nif_backend.ttl.

title

We should avoid serializing ontology iris in the import statement if possible. Currently we avoid this by not including the relevant prefixes, however if they creep in as an ns1 (e.g. http://purl.obolibrary.org/obo/) then the ontology iris will also be shortened, which is irritating from a usability standpoint.

http://purl.obolibrary.org/obo/SO_0000110

Due to the... large number of implicit conventions in neuron lang, many features fail to work because all en cell types does not follow the pattern those conventions require. Needs to be reworked to follow those convetions. @tmsincomb

Inconsistent types for return values cause headaches, switch from returning None to returning a list if requests returns None.

import rdflib
from pyontutils.core import makeGraph
g = makeGraph('buggraph', graph=rdflib.Graph())
g.g.add((rdflib.Literal('UT OH'), rdflib.RDFS.comment, rdflib.Literal('This will end badly.')))
g.write()

Produces a key error since subject literals don't seem to make it into the global sort keys.

The default turtle serializer produces the following.

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
@prefix xml: <http://www.w3.org/XML/1998/namespace> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

"UT OH" rdfs:comment "This will end badly." .

Has expression phenotype GABA but not neurotransmitter GABA. It should have the latter.

self._session.close() is being called more than once in some cases, a rare bug, but an annoying one

Also, I think we should be consistent about putting any spatial qualifiers after the main nucleus. We have a bunch of "medial nucleus X" and I think it should be "nucleus X, medial" where the spatial qualifier differentiates different parts of the nucleus.

We have an old class from NIF called Nucleic acid > RNA > mRNA > Neurofilament 150kD mRNA = http://uri.neuinfo.org/nif/nifstd/nlx_mol_090801

It is assigned to a few neuron classes. As it only has one subclass, I think this should probably be removed. Also given that it is a generic neuron protein, I'm not sure that encoding it for these classes is warranted, but if we are going to include inferred neuronal markers, then we should be consistent.

We have a class Morphological Quality with one member: http://uri.neuinfo.org/nif/nifstd/sao1278200674

I am assuming this should be deprecated in favor of the phenotype branch assigned to soma etc?

Replace: somatostatin with correct identifier
No soma location property recorded. Has layer locations but these are not reasoned over to determine that this is a neocortical cell.

When I run the reasoner, there are two classes for intrinsic neuron, one asserted and one equivalent, but each only has one neuron. We say that we classify according to this property and have indicated for each neuron the axon phenotype. So why don't they show up?

This is text/plain not application/json. May also want to upstream the issue.

Why do we have this class?

1. dematerialize needs to account for whether a relationship is transitive or not. For example for subClassOf it is correct to remove cases where a class occurs higher up, it is incorrect to do so for citations where the relationship is not transitive.
2. The asterisk marking the presence of additional parents is added too early/not removed when the graph is dematerialized.

still using NEMO identifiers... need to fix this

Use parcellation.py at 85404f5 and look at swanson_hierarchies.ttl to see the issue, orderings are not stable.

Running ttlcmp reveals that ttlser only deterministically serializes graphs that have identical BNode structures. There are a number of symmetric axioms in owl where OWLAPI will move a BNode from one parent class to another. We probably need to make ttlser minimally aware of cases where this can happen and come up with a consistent rule to prevent this from happening.

Hi,
I've followed the prerequisites steps in order to set up an env for NeuronLangExample.ipynb. The set up went well except for the following two operations:

  pip install dist/pyontutils-*-py3-none-any.whl

and

pip install dist/rdflib-*-py3-none-any.whl

that both finished with the following (non blocking) error:

Failed building wheel for mysql-connector

The notebook is launched at end but it is not working with the following error:
ModuleNotFoundError: No module named 'pyontutils'

I am using python 3.6.

Any chance you know what's wrong with this.

Only the first of 3 phenotypes is from the logical phenotype on layers is listed.

rdfs:label "Mammalia neocortex EGL (with-axon-in cortical layer I) Martinotti +GABA receptor role +Glutamate Receptor +SS +GABA interneuron"

NeuronCUT
and ((hasLayerLocationPhenotype some 'cortical layer II') or (hasLayerLocationPhenotype some 'cortical layer III') or (hasLayerLocationPhenotype some 'cortical layer V'))
and (hasCircuitRolePhenotype some 'Intrinsic Phenotype')
and (hasLocationPhenotype some neocortex)
and (hasMolecularPhenotype some somatostatin)
and (hasMolecularPhenotype some 'GABA receptor role')
and (hasMolecularPhenotype some 'Glutamate Receptor')
and (hasMorphologicalPhenotype some 'Martinotti Phenotype')
and (hasNeurotransmitterPhenotype some GABA)
and (hasTaxonRank some Mammalia)

http://uri.neuinfo.org/nif/nifstd/nifext_5068
and
http://purl.obolibrary.org/obo/PR_000005110

Both are used. Assuming we should deprecate the nifext class and substitute PRO

NLXWIKI:obi:IAO_0000088 -> http://neurolex.org/wiki/obi:IAO_0000088 on reserialization. It should not. Add tests to check for this. Only uris with # in the suffix should be disallowed since it is the comment literal for turtle.

This neuron lists both CR and Calb2 but I can find neither in table 7s for the MNC.

There seems to have been a change to something that joblib is doing. Possibly related to joblib/joblib#745. When running parcellation.py all pools show the following error.

Process ForkPoolWorker-1:
Traceback (most recent call last):
  File "/usr/lib64/python3.6/multiprocessing/process.py", line 258, in _bootstrap
    self.run()
  File "/usr/lib64/python3.6/multiprocessing/process.py", line 93, in run
    self._target(*self._args, **self._kwargs)
  File "/usr/lib64/python3.6/multiprocessing/pool.py", line 108, in worker
    task = get()
  File "/home/tom/.local/share/virtualenvs/pyontutils-a77fIBQE/lib/python3.6/site-packages/joblib/pool.py", line 149, in get
    return recv()
  File "/usr/lib64/python3.6/multiprocessing/connection.py", line 253, in recv
    return _ForkingPickler.loads(buf.getbuffer())
TypeError: __new__() takes 1 positional argument but 2 were given

So far this has been traced to /usr/lib64/python3.6/multiprocessing/connection.py when using backend='multiprocessing' but also occurs for other backends. When saving the pickle to disk and trying to reopen in another interpreter the same error occurs.

Need to fix how definedForTaxon is being used so that it doesn't act as an annotation property.

We have two equivalent classes: Mammalia neuron, one of which seems to contain the definition for a superior colliculus piriform neuron:
http://uri.neuinfo.org/nif/nifstd/BAMSC1117
NeuronCUT
and (hasLayerLocationPhenotype some 'Superior colliculus stratum opticum')
and (hasLocationPhenotype some 'Superior colliculus stratum opticum')
and (hasTaxonRank some Mammalia)
**

relationshipType appears to be being escaped correctly when it is an iri now. Not at all sure what changed but it means we are now double escaping things.

@CheerfulTeapot I think this is probably why has_proper_part was not working as expected?

In all other neurons, you use "hasSomaLocatedIn", but in this class, you have "hasLocationPhenotype" and "LayerLocationPhenotype". Please make it consistent. I cannot return the Martinotti CUT neuron, however, even with this query: Neuron and (hasLocationPhenotype some neocortex) and ((hasExpressionPhenotype some PR_000015665) or (hasExpressionPhenotype some Sst) or (hasExpressionPhenotype some somatostatin) or (hasExpressionPhenotype some Sst-IRES-Cre))

http://ontology.neuinfo.org/trees/query/rdfs:subClassOf/UBERON:0001948?flatten=true 500s

figure out how to do the replace for these since calling quote on them obviously wont work

Some lists do have implicit ordering in owl semantics and we are breaking that!

Need the conversion for SciGraph endpoint.

When loading a graph the debug.log file sticks around and creates noise in the zip file which should not be there.

http links to .org sites are being klobbered by this regex https://github.com/tgbugs/pyontutils/blob/master/pyontutils/docs.py#L253

Something like builder or similar, but need to come up with a better name. These depend on git and thus fall outside the core. Alternately, come up with a better version of makeGraph that covers the 80% use case and doesn't have all the baggage.

Similar issue in basic neurons with the swanson files.

Why do we have both cortical +SOM neuron and cortical somatostatin neuron?

Need to fix scigraph.py so that False, and True are subbed for the strings from the json (thought this had been fixed before...).

Should default to False, however keeping a local copy of results when you know your use case is likely to issue the same queries repeatedly could speed things up dramatically. Should probably has on the url or perhaps on the args.

push --force on a branch will cause ontload to fail silently if that branch is specified and the repo already exists

c = Config('output-file') seems like a better pattern for setting up export filename, imports, etc. We need to integrate core.Ont as well to get the prov related functionality as well.

There are many scripts that depend on resources and other repos and on being in a git repo which need to be separated from the core so that they can be tested independently. They should be distributed, but they need to be logically separate.