This repository contains classes required by Noctua/Minerva for representing entities that are object of 'enabled by' relations, and similar molecular relationships. This includes:
- genes
- protein (gene-level generic proteins and isoforms)
- functional RNAs
- complexes
These are represented as ontology classes, although NEO is not really an ontology in a conventional sense: there is no hierarchy, it is organized as a largely flat list. The purpose of distributing as an ontology is:
- Noctua is ontology-driven; curated create links between instances of classes
- RDF/OWL is the lingua-franca of the Noctua framework, and avoids the need for an ad-hoc format
- We can use reasoning to determine if relationships are valid
the GO Noctua instances loads the ontology go-lego.owl, which imports NEO
This GitHub repository only contains the tools required to build neo. The ontology is available from the following PURLS (Permanent URLs):
The build is handled by build-noctua-entity-ontology on Jenkins.
This runs the Makefile in this repository, and deploys the resulting ontology on S3, where it is available in multiple regions via cloudfront.
The contents of NEO are largely driven by the contents of each GOC member's GPI file (if a GPI file is not provided, a GAF is used instead). This allows each MOD or Database to have control over what kinds of entities can and cannot be described in Noctua.
The procedure for building NEO is relatively straightforward. The process is entirely automated, and no curation is involved. See the Makefile for details, but the overall procedure is:
- GOC datasets metadata is downloaded
- for each database contributing, the metadata is searched first for a GPI; if not present a GAF is used
- A simple script is executed converting the GAF or GPI into an ontology for that database/species
- The results are concatenated together into neo.obo / neo.owl
For modeling choices, we aim to be consistent with other ontologies
such as PRO. For example, the parent column in a GPI is used to make
has_gene_template relationships between the protein and its gene.
For genes we populate the UniProt Xref as a synonym to enhance autocomplete, and similarly gene IDs for proteins. This is to make it easier to select the correct entity type (typically a protein) for GOC members who are gene-centric.
Currently, the MGI GPI file uses PRO for protein entities, whereas the human GPI uses UniProtKB IDs. This means that a Noctua user can use PRO for mouse and UniProt for human.
See the issue tracker for full TODOs
RNAs come in via RNA central; we are still tweaking the pipeline, see issue tracker for details
Currently the OWL models produced by Noctua use gene entities (from
MGI, WB, etc) as the endpoint of enabled_by relationships. Note that
this is semantically incorrect, as this relationship type should be
used in conjunction with the molecule that has the activity, ie the
protein.
This was a short term decision to get us off the ground. Originally we
chose to interpret the MOD Gene ID X as the owl:unionOf (a) the
gene denoted by X (b) any gene product that is encoded_by some
X. However, this was found to be confusing and problematic.
Moving forward, the decision is to use the correct entity type at all
times. Thus the majority of the time the enabled_by will link to a
protein (or sometimes an ncRNA). One concern was that for MODs, it can
be difficult to select a protein ID that is guaranteed to permanently
have the desired semantics of "any product of gene X". To help, we:
- ensure that the gene ID is present as a synonym in the corresponding protein class, to facilitate accurate selection
- allow MODs control over what protein IDs are used via their GPI files. Thus if a MOD uses their own MOD protein IDs, these can be used. Alternatively the MOD can choose UniProt or PRO
At some stage we will switch out existing gene IDs for designated protein IDs.
The above applies to the scenario whereby the curator wants to describe activity for a generic product of a gene, and does not want to select a specific isoform (either because the function is believed to be held by all isoforms, or because isoform-level information is not known). Of course, when isoform level information is known, an isoform ID should be used. Again, this is under control of the contributing database via their GPI.
There are many subtleties here, but briefly:
- we use the UniProtKB GCPR entry to denote the generic entry (what PRO calls organism-gene-level)
- In a handful of cases, e.g. GNAS, there are multiple swissprot entries but only one GCPR. See below
NEO is generated automatically from GPIs, whereas PRO has a large curated component. However, in many cases they will have the same content. In particular, MGI provides lines in their GPI file that come from PRO, so we are in effect reconstituting PRO for the mouse subset.
PRO will largely overlap with UniProtKB. There are some subtle differences - see for example the representation of GNAS in human and mouse. Here guidelines may vary by database. For mouse, where PRO is used, the curator has access to precise semantics - either the organism-gene-level entry can be used, or a grouping isoform can be used.
Currently NEO builds are manually triggered.
Sometimes it can be difficult to figure out what's going on with the build. For those interested in understanding the build of a single resource target (e.g. mgi.gpi -> mgi-neo.obo), the following command can be useful:
make clean && TEST_SRCS=mgi make test_obo 2>&1 | tee /tmp/log.txt