INDRA (Integrated Network and Dynamical Reasoning Assembler) is an automated model assembly system interfacing with NLP systems and databases to collect knowledge, and through a process of assembly, produce causal graphs and dynamical models.
Home Page: http://indra.bio
License: BSD 2-Clause "Simplified" License
ActivityActivity statements are of the same form as the RasGef and RasGap queries. These queries should be collapsed into one, and the BelPy processor will get an instance of the appropriate BelPy statement based on the Activity of the object (if the object is a RasGTPase, then an increase means that the statement is about a RasGEF; if a decrease, then about a RasGAP).
For example, the protein name RAF1-BRAF is propagated through the PysbAssembler and into the PySB model, where it triggers an InvalidComponentNameError, e.g.
bp = biopax_api.process_pc_neighborhood(['BRAF'])
bp.get_phosphorylation()
pa.add_statements(bp.statements)
pa.make_model()
...
InvalidComponentNameError: Not a valid component name: 'RAF1-BRAF'
To use a test suite to measure our progress in implementing various features of the parser, assemblers, etc., it would be nice to have these tests separate from the tests of the correctness of basic functionality, which we want to run on Travis.
Query:
from indra.biopax import biopax_api as ba
bp = ba.process_pc_neighborhood(['MAP2K1', 'MAP2K2', 'MAPK1', 'MAPK3'])
bp.get_dephosphorylation()
This yields, among others, the statements:
Dephosphorylation(MAP2K1(), MAPK3(), PhosphorylationTyrosine, 203),
Dephosphorylation(MAP2K1(), MAPK3(), PhosphorylationThreonine, 201),
Dephosphorylation(MAP2K2(), MAPK3(), PhosphorylationTyrosine, 203),
Dephosphorylation(MAP2K2(), MAPK3(), PhosphorylationThreonine, 201),
Dephosphorylation(MAP2K1(), MAPK1(), PhosphorylationTyrosine, 184),
Dephosphorylation(MAP2K1(), MAPK1(), PhosphorylationThreonine, 182),
Dephosphorylation(MAP2K2(), MAPK1(), PhosphorylationTyrosine, 184),
Dephosphorylation(MAP2K2(), MAPK1(), PhosphorylationThreonine, 182),
Dephosphorylation(MAP2K1(), MAPK3(), PhosphorylationTyrosine, 203),
Dephosphorylation(MAP2K1(), MAPK3(), PhosphorylationThreonine, 201),
Dephosphorylation(MAP2K2(), MAPK3(), PhosphorylationTyrosine, 203),
Dephosphorylation(MAP2K2(), MAPK3(), PhosphorylationThreonine, 201),
Dephosphorylation(MAP2K1(), MAPK1(), PhosphorylationTyrosine, 184),
Dephosphorylation(MAP2K1(), MAPK1(), PhosphorylationThreonine, 182),
Dephosphorylation(MAP2K2(), MAPK1(), PhosphorylationTyrosine, 184),
Dephosphorylation(MAP2K2(), MAPK1(), PhosphorylationThreonine, 182),
Which appear to be artifactual. Strangely, they don't seem to turn up in the corresponding pathsbetween query among the same genes.
We should have a scripts folder at the top level folder, which contains scripts to create some of the resource files that are now under version control. This includes the different ID mappings: HGNC, UniProt, ChEBI, PubChem, and also the hierarchy RDFs.
After installing or cloning INDRA, the user could just run a script like
python scripts/get_resources.py
which would donwload/create the required resource files.
If you run the following script:
from indra.biopax import biopax_api as bpa
from indra.preassembler import Preassembler, render_stmt_graph
from indra.preassembler.hierarchy_manager import \
modification_hierarchy as mh, entity_hierarchy as eh
bpp = bpa.process_pc_pathsfromto(['MAP2K1'], ['MAPK1'])
bpp.get_phosphorylation()
bpp.get_activity_modification()
pa1 = Preassembler(eh, mh, bpp.statements)
pa1.combine_duplicates()
You get a number of statements including the following:
[ActivityModification(MAPK1(mods: (phosphorylation, Y, 184), (active), (phosphorylation, T, 182)), [(phosphorylation, T, 182), (phosphorylation, Y, 184)], increases, Activity),
ActivityModification(MAPK1(mods: (phosphorylation, Y, 184), (active), (phosphorylation, T, 182)), [(phosphorylation, Y, 184), (phosphorylation, T, 182)], increases, Activity),
ActivityModification(MAPK1(mods: (active), (phosphorylation, Y, 185), (phosphorylation, T, 183)), [(phosphorylation, Y, 185), (phosphorylation, T, 183)], increases, Activity),
ActivityModification(MAPK1(mods: (phosphorylation, Y, 187), (active), (phosphorylation, T, 185)), [(phosphorylation, T, 185), (phosphorylation, Y, 187)], increases, Activity),
ActivityModification(MAPK1(mods: (phosphorylation, Y, 187), (active), (phosphorylation, T, 185)), [(phosphorylation, Y, 187), (phosphorylation, T, 185)], increases, Activity),
ActivityModification(MAPK3(mods: (phosphorylation, Y, 203), (phosphorylation, T, 201), (active)), [(phosphorylation, T, 201), (phosphorylation, Y, 203)], increases, Activity),
ActivityModification(MAPK3(mods: (phosphorylation, Y, 203), (phosphorylation, T, 201), (active)), [(phosphorylation, Y, 203), (phosphorylation, T, 201)], increases, Activity),
ActivityModification(MAPK3(mods: (phosphorylation, T, 202), (active), (phosphorylation, Y, 204)), [(phosphorylation, T, 202), (phosphorylation, Y, 204)], increases, Activity),
These statements in many cases include the "active" flag on the agent, and in addition have the activating modifications as operating on the already modified protein. Should we change this to indicate that activating modification is relative to the unmodified protein?
This appears to be a problem only in the BEL API, not Biopax.
Dephosphorylation adds no new functionality on top of Modification, the only difference is that it has a field 'phos' instead of 'enz'. Changing this will require all code that touches Dephosphorylation to use the enz fieldname instead of phos.
We should implement a class called ModificationCondition similar to what we have for BoundCondition to represent Agent modifications. We will then be able to conveniently encode modification types, sites and negation flags.
As a simple use case for alternative assembly methods, consider the case where the purpose of assembly is only to produce a contact map. Here the goal is not to produce kinetics, but rather to provide a kind of graphical interactome of the relationships between the members. Here the rules produced by the assembly methods would be binding rules rather than, e.g., one-step conversion rules. Note however, that each statement might still need to specify its own assembly approach, since the binding relationships inferred by a Phosphorylation statement would likely limit all binding to a kinase active site, for example.
As a result, each assemble and monomers method would need to gain information on which approach to use. Since statement instantiation should be independent of assembly, this information should not be passed into each constructor at Statement instantiation time, but rather passed in as an argument to each assembly method. Thus perhaps a data structure at the level of the BelProcessor instance is most appropriate.
The data structure would likely be some kind of dict, with an entry for each statement type and a value indicating the type of assembly method to use. An additional possibility is for the simple specification of global settings, e.g., "contact map", "one-step conversion", etc. that would apply uniform policies for all statement types.
Perhaps this could start out as a dict and then, if necessary, graduate to a dedicated class.
Might be worthwhile having a hierarchy of activity types so that statements involving activities at various levels of resolution could also be subject to refinements.
For example, the statement 'X increases the kinase activity of Y' should be a refinement of 'X increases the activity of Y'.
Not sure how important this will be in practice--it's not clear how often the distinctions will be made in database entries or text.
Currently these are not implemented at all.
Currently the Biopax processor queries for statements involving complexes and then returns the results as a series of two-member Complexes. This should instead return Complex statements with the full list of members, indicating that the members can be part of the same connected component.
The Complex statement constructor takes a list of members (member names). We don't have a way of specifying that a species is in a modified form.
It would make sense to have a separate statement called Bind which describes direct binding through a site or domain. The current Complex([A, B, C]) statement means "A, B and C can be in a complex" or "There exists an A-B-C complex". From BioPax and BEL that is exactly how we extract complexes - we don't extract them from processes but rather from complex entities. Because of these semantics, Complex statements can naturally have multiple members even though molecules typically don't bind all at once. So historically, this makes sense.
But we should also be able to express the process that "The SH2 domain of GRB2 binds EGFR on a phosphorylated tyrosine residue". The main question is: how do we represent the binding sites? We could have 2 binding site arguments in the statement which could be strings.
This would work fine for "SH2" or "RBD" but not clear how to best do it for "phosphorylated tyrosine residue". I'm thinking of course that we ought to be able to then correctly PySB assemble this into EGFR(Y='p').
Consider the sentences:
"XXX results in phosphorylation of EGFR and tyrosine 1068 and tyrosine 1172."
"GRB2 binds EGFR phosphorylated at residue 1068."
In parsing the second sentence about GRB2, a site name is ultimately required for the modification being referred to as a precondition for the binding event. However, it is not explicitly specified in that sentence that that residue is a tyrosine, thus, while the first statement will result in a Phosphorylation sentence that leads to specification of a site with name "Y1068" on EGFR, the second statement will not have sufficient information to identify the name of the appropriate site.
This points to the need to have a aggregation/inference step at assembly time for these issues to be resolved. If we have information anywhere about what the residue at 1068 is, then we use it in all references to that site.
Similarly, we may see the statement "GRB2 binds phosphorylated EGFR." Again, at assembly time the binding statement should key this back to the appropriate site modifications. This one is more advanced because the assembly process would need to identify the issue that it is unresolved whether phospho 1068 or 1172 is sufficient, or both must be phosphorylated.
The query
bp = biopax_api.process_pc_pathsbetween(['MAP2K1', 'MAPK1'], neighbor_limit=1)
bp.get_phosphorylation()
returns
Phosphorylation(YWHAE, MAPK1, PhosphorylationThreonine, 182),
Phosphorylation(YWHAE, MAPK1, PhosphorylationTyrosine, 184),
which are TxY positions on mouse MAPK1 (http://www.uniprot.org/uniprot/P63085) whereas the TxY motif is at 185-187 on human MAPK1 (http://www.uniprot.org/uniprot/P28482).
Some modification positions are invalid (or are printed incorrectly):
Phosphorylation(MEK, ERK, Phosphorylation, -2147483648)
Typical output from Travis:
$ nosetests --with-doctest
EFSLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further details.
No output has been received in the last 10m0s, this potentially indicates a stalled build or something wrong with the build itself.
The build has been terminated
It could make sense to generalize ActivityModification to a statement called ActivityCondition. ActivityConditions could have an arbitrary mixture of ModCondition, BoundCondition and MutCondition conditions associated with them. We could then describe, for instance, that "KRAS bound to GTP is active". Or that "BRAF-V600E not bound to Vemurafenib is active". Does this make sense? If yes, it's an easy addition.
So far the corpora we've looked at don't have identically overlapping rules, though these are certainly a possibility that should be checked for. In particular these should be checked for after case normalization, to make sure that the same statement doesn't go in twice with the different capitalizations of the gene names (see issue #2).
This would be the first step in a more general process of comparing the content of statements for possible inconsistencies (e.g., "phosphoserine is activating" in conflict with "phosphoserine is inactivating").
In parsing statements in the BEL large corpus, we frequently find statements involving protein families, e.g. "AKT_FAMILY". Currently these enter the model as monomers named by the family. Instead, we should locate the mapping between the family and its specific members (and their HGNC names) and replace the statements involving families with all of the relevant family members.
When running the tests repeatedly, this line assert(pa.unique_stmts[0] == p5) # MEK phos ERK
sometimes fails and other times doesn't.
At some point policies became an argument in the constructor of the PysbAssembler. Wouldn't it make sense to put that back as an argument of make_model? The assembler collects statements and we would want to be able to generate multiple models from the same statements using the assembler by applying different policies.
Assume we have a processor called rp with len(rp.statements) > 0.
pa = Preassembler(eh, mh)
pa.add_statements(rp.statements)
At this point rp.statements is unchanged.
duplicate_stmts = pa.combine_duplicates()
At this point rp.statements has changed! This obviously shouldn't happen.
In processing the results from the query generated by the following script:
from indra.biopax import biopax_api as ba
bp = ba.process_pc_neighborhood(['TP63'])
bp.get_phosphorylation()
The following error occurs:
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further details.
Sending Pathway Commons query...
Pathway Commons query returned model...
indra/biopax/processor.py:251: UserWarning: Cannot handle complex enzymes.
warnings.warn('Cannot handle complex enzymes.')
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/p63_test.py in <module>()
2
3 bp = ba.process_pc_neighborhood(['TP63'])
----> 4 bp.get_phosphorylation()
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/biopax/processor.pyc in get_phosphorylation(self, force_contains)
53 def get_phosphorylation(self, force_contains=None):
54 stmts = self._get_generic_modification('phospho',
---> 55 force_contains=force_contains)
56 for s in stmts:
57 self.statements.append(Phosphorylation(*s))
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/biopax/processor.pyc in _get_generic_modification(self, mod_filter, mod_gain, force_contains)
274 enzs = BiopaxProcessor._get_agents_from_entity(controller_pe)
275 subs = BiopaxProcessor._get_agents_from_entity(input_spe,
--> 276 expand_pe=False)
277 for enz, sub in itertools.product(listify(enzs), listify(subs)):
278 # If neither the required enzyme nor the substrate is
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/biopax/processor.pyc in _get_agents_from_entity(bpe, expand_pe, expand_er)
413 # If it is a single entity, we get its name and database
414 # references
--> 415 name = BiopaxProcessor._get_element_name(bpe)
416 db_refs = BiopaxProcessor._get_db_refs(bpe)
417 agent = Agent(name, db_refs=db_refs, mods=mods, mod_sites=mod_sites)
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/biopax/processor.pyc in _get_element_name(bpe)
503
504 # Canonicalize name
--> 505 name = re.sub(r'[^\w]', '_', name)
506 if re.match('[0-9]', name) is not None:
507 name = 'p' + name
/Users/johnbachman/.virtualenvs/p27env/lib/python2.7/re.pyc in sub(pattern, repl, string, count, flags)
153 a callable, it's passed the match object and must return
154 a replacement string to be used."""
--> 155 return _compile(pattern, flags).sub(repl, string, count)
156
157 def subn(pattern, repl, string, count=0, flags=0):
TypeError: expected string or buffer
The immediate source of the error is that the string in the call to sub is None; however, the reason for this is that in the _get_element_name static method of the Biopax processor, the HGNC lookup fails. The HGNC ID reported by the call to _get_hgnc_id is 13473, but the subsequent call to _get_hgnc_name(hgnc_id) using that ID returns None, which is assigned to name, and is subsequently than passed to the regex.
The display name for the element causing the problem is "p63". A lookup at genenames.org indicates that the actual HGNC ID for TP63 is 15979.
Perhaps the right thing to do here is to simply check if the HGNC ID lookup fails, and if so, use the getDisplayName string instead?
We might want to keep a record of cases where the HGNC IDs have turned out to be erroneous so we can fix them.
I get the following phosphorylation statements from Pathway Commons:
Phosphorylation(EGFR(), SHC1(), PhosphorylationTyrosine, 194),
Phosphorylation(EGFR(), SHC1(), PhosphorylationTyrosine, 195),
Phosphorylation(EGFR(), SHC1(), PhosphorylationTyrosine, 272),
Phosphorylation(EGFR(), SHC3(), PhosphorylationTyrosine, 218),
Phosphorylation(EGFR(), SHC3(), PhosphorylationTyrosine, 219),
Phosphorylation(EGFR(), SHC3(), PhosphorylationTyrosine, 256),
Phosphorylation(EGFR(), SHC3(), PhosphorylationTyrosine, 257),
Phosphorylation(EGFR(), SHC3(), PhosphorylationTyrosine, 283),
Phosphorylation(EGFR(), SHC3(), PhosphorylationTyrosine, 301)]
These all raise errors against the Uniprot reference sequence. However, for SHC3, there is another isoform, the "p52" isoform, with a difference upstream sequence:
http://www.uniprot.org/uniprot/Q92529#sequences
In phosphosite it's clear that all of the tyrosines mentioned are correct for this isoform:
http://www.phosphosite.org/proteinAction.action?id=4936&showAllSites=true (click "show isoforms" in the chart below).
The same is true for SHC1, but this time it's isoform p46:
http://www.uniprot.org/uniprot/P29353#sequences
I also found the same to thing for RPS6KB1, where the T389 listed in Pathway Commons is correct, but for the "Isoform Alpha 2" which is missing the first 23 amino acids.
Is it possible that Pathway Commons has information on the isoforms involved in the statements and we are not extracting it?
If not, we could write check_statements to not only determine if the sequence is not a match, but also to see if it is a match for a different isoform of the protein.
An observation from the Ras 220 model. None of the Complex statements have evidence associated with them.
In [1]: from indra.bel import bel_api as ba
INFO:rdflib:RDFLib Version: 4.2.1
In [2]: bp = ba.process_ndex_neighborhood(['BRAF', 'MAP2K1'])
INFO:requests.packages.urllib3.connectionpool:Starting new HTTP connection (1): services.bigmech.ndexbio.org
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
http://www.openbel.org/vocabulary/Phosphorylation
Getting all direct statements...
Checking for 'degenerate' statements...
Protein -> Protein/Activity statements:
---------------------------------------
Total indirect statements: 166
Total direct statements: 40
Converted statements: 36
Degenerate statements: 0
>> Total unhandled statements: 4
--- Unhandled statements ---------
complex_p_HGNC_BRAF_p_HGNC_PRKCE_p_HGNC_RPS6KB2_DirectlyIncreases_kin_p_HGNC_PRKCE
p_HGNC_BRAF_sub_V_600_E_DirectlyIncreases_kin_p_HGNC_BRAF
a_CHEBI_sorafenib_tosylate_DirectlyDecreases_kin_p_HGNC_BRAF
complex_p_HGNC_BRAF_p_HGNC_RAF1_DirectlyIncreases_kin_p_HGNC_RAF1
--- Converted BelPy Statements -------------
0: Complex(['EIF4G1', 'EIF4E'], [])
1: Complex(['BRAF', 'RAF1'], [])
2: Complex(['MAP2K1', 'PTPN7'], [])
3: Complex(['RPS6KB2', 'PRKCE', 'BRAF'], [])
4: Complex(['MAP2K1', 'GIT1'], [])
5: Complex(['MAP2K1', 'RAF1'], [])
6: Phosphorylation(PRKA_FAMILY, BRAF, Phosphorylation, 429, [Evidence(bel, http://bio2rdf.org/pubmed:15212693, [], 11510412;10869359)])
7: Phosphorylation(AKT_FAMILY, BRAF, Phosphorylation, 429, [Evidence(bel, http://bio2rdf.org/pubmed:15212693, [], 11510412;10869359)])
8: Phosphorylation(MAP2K1, MAPK3, Phosphorylation, 202, [Evidence(bel, http://bio2rdf.org/pubmed:15212693, [], 8626767;93330262;15466476;11971971)])
9: Phosphorylation(RAF1, MAP2K1, Phosphorylation, 218, [Evidence(bel, http://bio2rdf.org/pubmed:12937126, [], MEK1/2 is activated through phosphorylation at Ser 217/Ser 221 catalyzed by Raf.)])
---------------------------------------------------------------------------
UnicodeEncodeError Traceback (most recent call last)
<ipython-input-2-a69375dadcfa> in <module>()
----> 1 bp = ba.process_ndex_neighborhood(['BRAF', 'MAP2K1'])
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/bel/bel_api.py in process_ndex_neighborhood(gene_names, rdf_out)
13 with open(rdf_out, 'wt') as fh:
14 fh.write(rdf.encode('utf-8'))
---> 15 bp = process_belrdf(rdf)
16 bp.print_statements()
17 return bp
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/bel/bel_api.py in process_belrdf(rdf_str)
35 bp.print_statement_coverage()
36 print "\n--- Converted BelPy Statements -------------"
---> 37 bp.print_statements()
38 return bp
39
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/bel/processor.py in print_statements(self)
662 def print_statements(self):
663 for i, stmt in enumerate(self.statements):
--> 664 print "%s: %s" % (i, stmt)
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/statements.py in __str__(self)
215 s = ("%s(%s, %s, %s, %s, %s)" %
216 (type(self).__name__, self.enz.name, self.sub.name, self.mod,
--> 217 self.mod_pos, self.evidence))
218 return s
219
UnicodeEncodeError: 'ascii' codec can't encode character u'\u2013' in position 127: ordinal not in range(128)
Here's a statement I came across from Biopax:
ActivityModification(MAP2K1(mods: ['PhosphorylationSerine'], mod_sites: [298]),
['PhosphorylationSerine', 'PhosphorylationSerine', 'PhosphorylationSerine'], ['218', '298', '222'],
increases, Activity)
MAP2K1 is listed as starting out as being phosphorylated at serine 298, but the activity modification includes that site again as being activating. Presumably this results from an error in excluding serine 298 from the modification list when examining the BioPax reaction.
Activity flags are still generated into monomer signatures, even if they don't appear in any rules.
E.g. all rules where MAP2K1 is active, it is represented as
MAP2K1(S218='p',S222='p'),
but the monomer has the signature
Monomer(u'MAP2K1', ['S218', 'S222', 'Kinase', 'Active'], {'Active': ['inactive', 'active'], 'S218': ['u', 'p'], 'Kinase': ['inactive', 'active'], 'S222': ['u', 'p']})
Currently the BEL processor ignores statements about protein modifications where the left-hand side (enzyme term) is the abundance of a complex (rather than a protein abundance or the activity of a protein abundance). This should be appropriately parsed into a BEL term with the agent having the bound conditions indicating the members of the complex.
However, one challenge of doing this is that the INDRA statement representation currently requires one of the proteins to be indicated as the enzyme, whereas BEL statements involving complexes do not indicate which protein is the enzyme responsible for the modification.
======================================================================
FAIL: test_trips.test_phosphorylation
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/travis/miniconda/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/home/travis/build/johnbachman/indra/indra/tests/test_trips.py", line 12, in test_phosphorylation
assert(len(tp.statements) == 1)
AssertionError:
-------------------- >> begin captured stdout << ---------------------
All events by type
------------------
------------------
--------------------- >> end captured stdout << ----------------------
======================================================================
FAIL: test_trips.test_phosphorylation_noresidue
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/travis/miniconda/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/home/travis/build/johnbachman/indra/indra/tests/test_trips.py", line 20, in test_phosphorylation_noresidue
assert(len(tp.statements) == 1)
AssertionError:
-------------------- >> begin captured stdout << ---------------------
All events by type
------------------
------------------
--------------------- >> end captured stdout << ----------------------
======================================================================
FAIL: test_trips.test_phosphorylation_nosite
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/travis/miniconda/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/home/travis/build/johnbachman/indra/indra/tests/test_trips.py", line 28, in test_phosphorylation_nosite
assert(len(tp.statements) == 1)
AssertionError:
-------------------- >> begin captured stdout << ---------------------
All events by type
------------------
------------------
--------------------- >> end captured stdout << ----------------------
======================================================================
FAIL: test_trips.test_actmod
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/travis/miniconda/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/home/travis/build/johnbachman/indra/indra/tests/test_trips.py", line 36, in test_actmod
assert(len(tp.statements) == 1)
AssertionError:
-------------------- >> begin captured stdout << ---------------------
All events by type
------------------
------------------
--------------------- >> end captured stdout << ----------------------
======================================================================
FAIL: test_trips.test_actmods
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/travis/miniconda/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
self.test(*self.arg)
File "/home/travis/build/johnbachman/indra/indra/tests/test_trips.py", line 57, in test_actmods
assert(len(tp.statements) == 1)
AssertionError:
-------------------- >> begin captured stdout << ---------------------
All events by type
------------------
------------------
--------------------- >> end captured stdout << ----------------------
----------------------------------------------------------------------
Ran 252 tests in 43.860s
FAILED (failures=5)
Example::
In [16]: bp = biopax_api.process_pc_pathsfromto(['KSR1'], ['BRAF'])
In [17]: bp.get_phosphorylation()
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/biopax/processor.py:164: UserWarning: Unknown modification type residue modification, inactive
warnings.warn('Unknown modification type %s' % mf_type)
In [18]: bp.statements
Out[18]:
[Phosphorylation(KSR1, BRAF, PhosphorylationSerine, 151, [Evidence(biopax, [], [], None)]),
Phosphorylation(KSR1, BRAF, PhosphorylationSerine, 753, [Evidence(biopax, [], [], None)]),
Phosphorylation(KSR1, BRAF, PhosphorylationSerine, 578, [Evidence(biopax, [], [], None)]),
Phosphorylation(KSR1, BRAF, PhosphorylationSerine, 750, [Evidence(biopax, [], [], None)]),
Phosphorylation(KSR1, BRAF, PhosphorylationThreonine, 401, [Evidence(biopax, [], [], None)])]
This makes it appear that KSR1 is the kinase involved in the phosphorylation. Is this because KSR1 appears as part of a complex on the left hand side but is not itself the kinase? Can we disambiguate which one is the kinase?
From Pathway Commons:
RAS_family
MKK4_MKK7
RAF1_BRAF
MEK1_2
MEK1_2_active
Raf
RAC1_CDC42
MAPK
p14_3_3_family
Queries of BEL or BioPax can (and do) result in multiple equivalent statements, as in the following. They should be handled in some way, either by incrementing the count, or perhaps adding to a list of papers/references/contexts or other information that may differ between the different instances.
bp = biopax_api.process_pc_neighborhood(['BRAF'])
bp.get_phosphorylation()
...
Phosphorylation(NRAS, RAF1, PhosphorylationThreonine, 268)
Phosphorylation(NRAS, RAF1, PhosphorylationThreonine, 268)
We can currently get full text for any PMC OAI paper using the PMC client. We can only get abstracts for papers from the PubMed client. The Elsevier client allows us to access full text based on DOI for papers under the publisher. It would be good to write a top level wrapper that can do ID mapping (PMCID - PMID - DOI) and determine if the full text is accessible through some resource and get the full text through that resource.
Some of the statements in the groups below should have been preassembled out.
Phosphorylation(MAP2K1(mods: (phosphorylation), (phosphorylation)), MAPK1()),
Phosphorylation(MAP2K1(mods: (phosphorylation, S, 218), (phosphorylation, S, 298), (phosphorylation, S, 222)), MAPK1(), T, 185),
Phosphorylation(MAP2K1(mods: (phosphorylation, S, 218), (phosphorylation, S, 298), (phosphorylation, S, 222)), MAPK1(), Y, 187),
Phosphorylation(RAF1(), MAP2K1(mods: (phosphorylation, S, 298)), S, 218),
Phosphorylation(RAF1(), MAP2K1(mods: (phosphorylation, S, 298)), S, 222),
Phosphorylation(RAF1(mods: (phosphorylation), (phosphorylation), (phosphorylation)), MAP2K1()),
Phosphorylation(RAF1(mods: (phosphorylation), (phosphorylation)), MAP2K1()),
Phosphorylation(RAF1(mods: (phosphorylation, S, 296), (phosphorylation, T, 269), (phosphorylation, S, 471), (phosphorylation, S, 289), (phosphorylation, S, 338), (phosphorylation, S, 301)), MAP2K1(), S, 218),
Phosphorylation(RAF1(mods: (phosphorylation, S, 296), (phosphorylation, T, 269), (phosphorylation, S, 471), (phosphorylation, S, 289), (phosphorylation, S, 338), (phosphorylation, S, 301)), MAP2K1(), S, 222),
Phosphorylation(RAF1(mods: (phosphorylation, T, 269), (phosphorylation, S, 621), (phosphorylation, S, 471), (phosphorylation, S, 338)), MAP2K1(), S, 218),
Phosphorylation(RAF1(mods: (phosphorylation, T, 269), (phosphorylation, S, 621), (phosphorylation, S, 471), (phosphorylation, S, 338)), MAP2K1(), S, 222),
Phosphorylation(RAF1(mods: (phosphorylation, S, 621), (phosphorylation, T, 268)), MAP2K1(), S, 218),
Phosphorylation(RAF1(mods: (phosphorylation, S, 621), (phosphorylation, T, 268)), MAP2K1(), S, 222),
Phosphorylation(RAF1(mods: (phosphorylation, S, 621), (phosphorylation, T, 268)), MAP2K1(), T, 286),
Phosphorylation(RAF1(mods: (phosphorylation, S, 621), (phosphorylation, T, 268)), MAP2K1(), T, 292),
Phosphorylation(RAF1(mods: (phosphorylation, S, 621), (phosphorylation, T, 268)), MAP2K1(), T, 386),
Phosphorylation(RAF1(mods: (phosphorylation, T, 269), (phosphorylation, T, 268)), MAP2K1(), S, 218),
Phosphorylation(RAF1(mods: (phosphorylation, T, 269), (phosphorylation, T, 268)), MAP2K1(), S, 222),
Phosphorylation(RAF1(mods: (phosphorylation, S, 338)), MAPK1(), T, 185),
Phosphorylation(RAF1(mods: (phosphorylation, S, 338)), MAPK1(), Y, 187),
I'm going to use this issue to keep an ongoing record of suspicious statements pulled out of Pathway Commons. Many (most, all?) of these may be either errors or deliberate curation shortcuts put in the database itself, but I'll list these as an issue in case they are appearing because we are doing something incorrectly in the api.
GTP as a kinase:
Phosphorylation(GTP, MAPK10, PhosphorylationThreonine, 221),
Phosphorylation(GTP, MAPK10, PhosphorylationTyrosine, 223),
Phosphorylation(GTP, MAPK8, PhosphorylationThreonine, 183),
Phosphorylation(GTP, MAPK8, PhosphorylationTyrosine, 185),
Phosphorylation(GTP, MAPK9, PhosphorylationThreonine, 183),
Phosphorylation(GTP, MAPK9, PhosphorylationTyrosine, 185),
H, K, and NRAS as kinases (more understandable):
Phosphorylation(HRAS, BRAF, PhosphorylationSerine, 578),
Phosphorylation(HRAS, BRAF, PhosphorylationThreonine, 373),
Phosphorylation(HRAS, RAF1, PhosphorylationSerine, 338),
Phosphorylation(HRAS, RAF1, PhosphorylationSerine, 621),
Phosphorylation(HRAS, RAF1, PhosphorylationThreonine, 268),
Phosphorylation(HRAS, RAF1, PhosphorylationThreonine, 269),
Phosphorylation(KRAS, BRAF, PhosphorylationSerine, 578),
Phosphorylation(KRAS, BRAF, PhosphorylationThreonine, 373),
Phosphorylation(KRAS, RAF1, PhosphorylationSerine, 621),
Phosphorylation(KRAS, RAF1, PhosphorylationThreonine, 268),
Phosphorylation(KRAS, RAF1, PhosphorylationThreonine, 269),
Phosphorylation(NRAS, BRAF, PhosphorylationSerine, 578),
Phosphorylation(NRAS, BRAF, PhosphorylationThreonine, 373),
Phosphorylation(NRAS, RAF1, PhosphorylationSerine, 621),
Phosphorylation(NRAS, RAF1, PhosphorylationThreonine, 268),
Phosphorylation(NRAS, RAF1, PhosphorylationThreonine, 269),
IL2 (ligand) as a kinase (perhaps this is meant as indirect?):
Phosphorylation(IL2, MAPK1, PhosphorylationThreonine, 182),
Phosphorylation(IL2, MAPK1, PhosphorylationTyrosine, 184),
Phosphorylation(IL2, MAPK3, PhosphorylationThreonine, 201),
Phosphorylation(IL2, MAPK3, PhosphorylationTyrosine, 203),
KSR1 as a kinase:
Phosphorylation(KSR1, MAPK1, PhosphorylationThreonine, 182),
Phosphorylation(KSR1, MAPK1, PhosphorylationTyrosine, 184),
Phosphorylation(KSR1, MAPK3, PhosphorylationThreonine, 201),
Phosphorylation(KSR1, MAPK3, PhosphorylationTyrosine, 203),
14-3-3 proteins as kinases. Could it be these are involved in some kind of reaction describing negative regulation that's not getting parsed correctly?
Phosphorylation(YWHAB, MAPK1, PhosphorylationThreonine, 182),
Phosphorylation(YWHAB, MAPK1, PhosphorylationTyrosine, 184),
Phosphorylation(YWHAB, MAPK3, PhosphorylationThreonine, 201),
Phosphorylation(YWHAB, MAPK3, PhosphorylationTyrosine, 203),
Phosphorylation(YWHAE, MAPK1, PhosphorylationThreonine, 182),
Phosphorylation(YWHAE, MAPK1, PhosphorylationTyrosine, 184),
...
Phosphorylation(p14_3_3_family, MAPK1, PhosphorylationThreonine, 182),
Phosphorylation(p14_3_3_family, MAPK1, PhosphorylationTyrosine, 184),
Phosphorylation(p14_3_3_family, MAPK3, PhosphorylationThreonine, 201),
Phosphorylation(p14_3_3_family, MAPK3, PhosphorylationTyrosine, 203),
Inconsistency in sequence numbering for MAPK1. Also note that these two statements are not only inconsistent with each other, but also with the phosphorylation statements shown above for MAPK1, which lists the residues 182 and 184:
ActivityModification(MAPK1, ['PhosphorylationThreonine', 'PhosphorylationTyrosine'], [183, 185], increases, Activity),
ActivityModification(MAPK1, ['PhosphorylationThreonine', 'PhosphorylationTyrosine'], [185, 187], increases, Activity),
RAS mediates dephosphorylation of RAF1?
Dephosphorylation(HRAS, RAF1, PhosphorylationSerine, 621),
Dephosphorylation(KRAS, RAF1, PhosphorylationSerine, 621),
Dephosphorylation(NRAS, RAF1, PhosphorylationSerine, 621),
Dephosphorylation(RAS_family, RAF1, PhosphorylationSerine, 621),
Currently amino acids and their various abbreviations e.g. serine, ser, S are used as strings with no restrictions. It would be better to have a pre-defined dictionary with all amino acid names and their abbreviations in statements.py.
In processors: if an amino-acid name in a new Statement that a processor constructs is invalid, a warning or error could be thrown from statements.py.
In assemblers: All assemblers should refer to amino acid abbreviations in statements.py and not use their own copy of the same lookup table.
Currently the links between agents are directed, presumably by the order of the agents in the Complex statement.
It appears that the query for complexes is not picking up all protein members, leading to Complex statements that contain only one member. Anecdotally it appears that the problem may be when one member of the complex is a ModifiedProteinAbundance rather than a ProteinAbundance.
I started implementing MutConditions - a named tuple to represent amino acid substitutions on INDRA Agents. I suggest the following format:
MutCondition = namedtuple('MutCondition', ['pos', 'residue_from', 'residue_to'])
There are several important questions that we need to consider. For instance:
When attempting to process the rdf from data/large_corpus.rdf:
In [5]: bp = bel_api.process_belrdf(rdf)
---------------------------------------------------------------------------
InvalidNameError Traceback (most recent call last)
<ipython-input-5-d77cea19fd57> in <module>()
----> 1 bp = bel_api.process_belrdf(rdf)
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/bel/bel_api.py in process_belrdf(rdf_str)
27 # Build BelPy statements from RDF
28 bp = BelProcessor(g)
---> 29 bp.get_complexes()
30 bp.get_activating_subs()
31 bp.get_modifications()
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/bel/processor.py in get_complexes(self)
350 for stmt in res_cmplx:
351 cmplx_name = term_from_uri(stmt[0])
--> 352 child_name = gene_name_from_uri(stmt[1])
353 child = Agent(child_name)
354 cmplx_dict[cmplx_name].append(child)
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/bel/processor.py in gene_name_from_uri(uri)
53
54 def gene_name_from_uri(uri):
---> 55 return name_from_uri(uri).upper()
56
57 def term_from_uri(uri):
/Users/johnbachman/Dropbox/1johndata/Knowledge File/Biology/Research/Big Mechanism/indra/indra/bel/processor.py in name_from_uri(uri)
48 pass
49 else:
---> 50 raise InvalidNameError(name)
51
52 return name
InvalidNameError: Not a valid name: Phosphatidylinositol_3,4,5_trisphosphate
Currently the processor fills in a few fields of a single Evidence object, whereas in reality a single biopax reaction might have a large number of references. The list of evidence information returned from biopax should be converted into a list of Evidence objects that are added to the statement.
ActiveForm(MAPK1(mods: (phosphorylation, T, 185)), kinase, True),
ActiveForm(MAPK1(mods: (phosphorylation, Y, 185)), kinase, True),
Gene names are parsed in from BELscript/RDF using whatever case they have, which leads to different rules having, e.g., "Braf" and others "BRAF". When converted into PySB, these lead to different monomers.
Possible solutions:
Here are two Statements with their matches_keys:
Complex(CCND1(), CDK4())
"(<class 'indra.statements.Complex'>, ("('CCND1', [], [], None, 0, ())", "('CDK4', [], [], None, 0, ())"))"
Complex(CCND1(), CDK4())
"(<class 'indra.statements.Complex'>, ("(u'CCND1', [], [], None, 0, ())", "(u'CDK4', [], [], None, 0, ())"))"
The only difference is that in one case the member names are unicode and in the other case they aren't. This breaks the logic of finding duplicates by matching by matches_key strings.
There appear to be many reactions in BioPax of the form
From: MEK()
Catalyzed by: Raf(Active state,R)
To: MEK(Active state,R)
i.e., an ActivityActivity statement involving Raf and Mek. We should be able to process these in BioPax as we do for BEL.
The ActivityActivity statements may prove to be particularly useful when assembling them alongside other statements where the identity of the modifying enzyme is actually not clear, for example when the reaction is catalyzed by a complex (e.g., 14-3-3 % BRAF % KSR).
We keep track of evidence at the level of Statements but the generated model itself is not annotated. We could annotate each PySB rule by, for instance, the natural language sentence, the BioPax reaction, or the BEL statement that it was processed from.
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.