The michelanglo_protein module collects all the data needed for analysing variants.
p = ProteinAnalyser(uniprot = 'Q86V25').load()
print(p)
p.mutation = Mutation('p.N127W')
p.analyse_structure()
print(p.get_features_near_position())
print(p.get_gnomAD_near_position())
print(p.model.get_structure_neighbours())
print(p.get_superficiality())
The data loaded is either gatherered from various databases, some of which need splitting (vide infra or create.py).
If it is just the one gene, you can use the following, which will retrieve the Uniprot data of the one gene (failing to get external data if unavailable):
p = ProteinGatherer(uniprot='Q86V25')
p.parse_uniprot()
The site michelanglo.sgc.ox.ac.uk depends on three repos:
- MichelaNGLo-app
- MichelaNGLo-transpiler
- MichelaNGLo-protein-module
The top-level repo is Michelanglo-and-Venus
This module can be used independently of Michelanglo app module, but requires the transpiler module.
If you are interested in the gene/protein name synonyms to uniprot mapping files see Name synomyms to Uniprot.
If you want human protein and don't want to generate the data (i.e. by running create.py) see human protein.
For installation instructions see relevant sections in Michelanglo deployment.
The michelanglo_protein module contains a .generate subfolder.
For speed at time of a request the module preparses a lot of things thanks to the .generate submodule.
Initially the Uniprot data was fetched online and parsed on the fly —technically this should still be possible with a few tweaks.
There are a few "protein" classes.
- These have different roles, but are all based off
ProteinCoreclass. - The parser protein class from
.generateisProteinGatherer, this has many fetching parts that are fault tolerant (changeable in settings). - For analysis of a mutation
ProteinAnalyseris used.
Additionally, there are classes whose instances are bound to these:
.settings, instance ofGlobalSettings, a singleton class, which controls the settings and folders and actually fetches the initial data (see below).Variantinstances appear in lists. gnomAD etc.- whereas
ProteinAnalyser.mutationis an instance ofMutation Structureinstances appear in lists. Stores crystal data.StructureAnalyseris a special case of the above forProteinAnalyser.structure
GlobalSettings appears as .settings of various classes. It is a singleton class, so the changes are global.
It is instatiated in settings_handler.py, but it does not create folders etc. until the .startup method is called (directly or indirectly by attempting anything).
global_settings.verbose = True False
global_settings.startup(data_folder='My-data')
global_settings.retrieve_references(ask=False, refresh=False)
ProteinGatherer is so big it is split across three files, it's mixin base classes are in protein._protein_uniprot_mixin (handles uniprot) and protein._protein_base_mixin.py (magic methods except __init__).
The _UniprotMixin requires Element-tree to be monkeypatched, which is done in ET_monkeypatch.
>>> from ET_monkeypatch import ET
The where-is-what logistics and other settings is stored in Protein.settings, which is an instance of protein.settings_handler.GlobalSettings().
>>> from protein.generate import ProteinGatherer as Protein
>>> Protein.settings.startup('temp_just_for_today') # create the folders.
>>> Protein.settings.verbose = True
>>> Protein.settings.missing_attribute_tolerant = True
>>> Protein(uniprot = 'Q9NWZ3').foo
warning... but does not die.
>>> Protein.settings.error_tolerant = True ##see the decorator failsafe for more.
>>> Protein(uniprot = 'foo').parse_uniprot() #will fail (extreme case)
warning... but does not die
>>> Protein.settings.data_folder = 'new_location' ## all subfolders will be created too.
For where the error tolerance gets applied see the mixin protein._protein_base_mixin._BaseMixin
The settings does not create on import the folders, but when the .init(data_folder='data') method is called or any operations are performed.
To change the folder of data midway use .data_folder = xxx, don't reinitialise.
This script used to be called Variant and generated static html pages (cf. HIFC2-Tracker_variant_classifier repo). It still can (maybe).
The pages folder is .page_folder and .wipe_html() clears them.
To change species see ET_monkeypatch.ET.is_human().
The data is a big. It needs slimming as unfortunately it has gotten bloated. So it needs tidying.
The module protein.generate creates the needed files.
At some point I started writing settings.retrieve_references() but stopped as it would have been for the once and some reference require passwords and some change faster than stairways in Hogwarts (ie. NCBI).
It mostly works. But downloading large datasets such as gnomAD will run out of RAM as it currently is not streamed.
Also as of December 2019 there is no exome gnomAD v3. So it has to be made from the gargantuan dataset.
NB. The script without the data will fetch off the web the Uniprot (only .parse_uniprot() will work without screaming).
See protein/settings_handler the class attributeGlobalSettings.addresses (or Protein.settings.addresses).
The offset of a PDB chain in the coordinate resi relative to Uniprot is based on SIFT data. Mostly.
In some cases there is both a missing N and C terminus (see offset_issue.png), which means that there is no PDB_BEG or PDB_END.
This is where the bound method of Structure .get_offset_from_PDB(detail, sequence) comes in. This is a fix to an already generated file.
It requires the sequence of the chain in question.
See also: blog post.
See Notes
For what is needed for VENUS (SNV analyser see that)
The uniprot parser part of this module could be integrated with Biopython. Currently there is no Uniprot parsing capability for it.
Sphinx documentation: