y1zhou/metabolike

Identifying reprogrammed metabolic routes given omics data.

Welcome to metabolike 👋

An alternative way to explore metabolic networks.

Metabolike lets you transform SBML metabolic models into queryable, interactive graphs.

Installation

pip install -U metabolike

For more details, please visit the documentation site.

Feature highlights

• Parser that can read any valid SBML model and write relevant entities into a graph database.
• Special support for BRENDA and BioCyc data files.
• A Streamlit app for novel metabolic route detection with omics-data integration.

License

Metabolike is completely free and open-source and licensed under the GPL-3.0 license.

metabolike's Issues

Information about citations in `pubs.dat`

• Map ID to UNIQUE-ID in the file; PUB- is prefixed
• List attributes: AUTHORS
• String attributes: DOI-ID, MEDLINE-UID, PUBMED-ID, REFERENT-FRAME, SOURCE, TITLE, URL, YEAR

Link BRENDA enzymes to the graph

Using EC numbers should be the simplest method.

Initial Update

The bot created this issue to inform you that pyup.io has been set up on this repo.
Once you have closed it, the bot will open pull requests for updates as soon as they are available.

Documentation

Currently all documentation are automatically-generated from the docstrings. The project should have standalone docs written to describe the workflow.

`classes.dat` contains the common name for a lot of the frames and IDs used in `pathways.dat`.

Anything not human-readable should have a corresponding UNIQUE-ID in classes.dat. For example, CCO-SURFACE-MAT is the cell surface matrix of CCO, and CCO is Cell Component Ontology.

• Cellular compartment CCO
• Common name for taxon, i.e. Organisms
• Evidence code in citations Evidence. The mcId for Citation nodes is sometimes :-delimited and has this part (related to #15)
• Gene-Ontology-Terms

Some reaction nodes are getting dropped

Related to #18 and c495b71. Unfortunately some reactions don't have a canonical ID in their multiple duplicates. For example, FUMHYDR-RXN[CCO-CYTOSOL]-MAL//FUM/WATER.28. and FUMHYDR-RXN[CCO-CYTOSOL]-MAL//FUM/WATER.28. both exist for humans, but FUMHYDR-RXN doesn't exist.

Link MetaCyc pathways to reactions

Unsure of the difference between reactions.dat and #reactions.dat#.

• #8
• Use pathways.dat to create annotation on Pathway nodes.
  • UNIQUE-ID are identifiers of the pathways, i.e. the BioCyc ID.
  • CITATIONS can have multiple entries and point to pubs.dat or Pathway Tools evidence ontology.
  • #11
  • IN-PATHWAY points to SuperPathway nodes. Seems to be the same with SUPER-PATHWAYS. Similarly for super pathways there's an SUB-PATHWAY attribute.
  • PATHWAY-LINKS are tuples with the first element being metabolites in the pathway, and other elements being IDs of other pathways.
  • PREDECESSORS connects reactions in the pathway. 2nd (and 3rd, etc.) element is the predecessor of the 1st element.
  • PRIMARY-PRODUCTS and PRIMARY-REACTANTS are beginning and ending metabolites.
  • REACTION-LAYOUT gives the primary metabolites and direction of each reaction.
  • REACTION-LIST enumerates the reaction IDs in the pathway.
  • SPECIES enumerates taxa known to possess this pathway.
  • SYNONYMS are names for the pathway.
  • TAXONOMIC-RANGE are the expected taxonomic range.
• #13

Parse links in the `COMMENT` field

COMMENT corresponds to the "Summary" section on the website, and often spans multiple lines containing links to various other attributes, e.g. |CITS:[Hill99]|. Once all the other nodes are created, there should be a function that parses the COMMENT attribute and create relationships with other nodes.

Weight edges in route search

Now the graph search can only identify shortest paths, but to be useful we need to weigh the edges based on omics data (gene expression, deleterious mutation, etc.) and thermodynamics information (Gibbs free energy change of reaction).

Add atom mappings to reactions and compounds

SMILES atom mappings in atom-mappings-smiles.dat and could be added to Reaction nodes.

Annotation of compounds

Compound nodes with (c:Compound)-[:is]->(rdf:RDF) relationships can be mapped to items in the compounds.dat file using:

• rdf.Biocyc to file UNIQUE-ID (strip the META: prefix).
• rdf.Inchikey to file INCHI-KEY (strip InChIKey= prefix in file).
• Other identifiers in RDF to DBLINKS in file; this is harder to parse and should only be considered when the other two don't give matches.

Once we have a match, the following information could be added:

• c node: GIBBS-0, LOGP, MOLECULAR-WEIGHT, MONOISOTOPIC-MW, POLAR-SURFACE-AREA, SYNONYMS, PKA[123], COMMENT
• rdf node: SMILES, INCHI, DBLINKS
• Links to other nodes: CITATIONS

Improve BRENDA load time

Using Dask to parallelize the Lark tree generation, but the multi-threaded client of map_partitions doesn't seem to speed up anything. Consider using multiple processes (simply switching causes weird bugs), or using @delayed function calls.

Duplicated canonical reactions

Some reactions have identical canonical reaction IDs, e.g. PGLUCISOM-RXN and PGLUCISOM-RXN-ALPHA-GLC-6-P//FRUCTOSE-6P.27.. This is caused by metabolic reactions with annotated substrates and/or cellular compartments in their IDs.

These are not the same reactions, but during the annotation step (

metabolike/metabolike/parser/metacyc.py

Line 335 in 80500ef

def reaction_to_graph(self, rxn_id: str, rxn_dat: Dict[str, List[List[str]]]):

) the two would get exactly the same information added. Is this the right thing to do?

Tests for neo4j-related code

Is there a way to mock the neo4j database? If not then how can the database-related code be tested?

Super-pathway nodes are not annotated

When parsing pathways.dat, the IN-PATHWAY and SUPER-PATHWAYS attributes may generate new pathway nodes that are not further annotated.

Reaction ID mismatch

Some reactions have longer ID forms in metabolic-reactions.xml than in reactions.dat or pathways.dat. For example, F16BDEPHOS-RXN has two counterparts in metabolic-reactions.xml:

[x for x in all_rxns if x.startswith("F16BDEPHOS-RXN")]
# [
#    'F16BDEPHOS-RXN[CCO-PERI-BAC]-FRUCTOSE-16-DIPHOSPHATE/WATER//FRUCTOSE-6P/Pi.60.',
#    'F16BDEPHOS-RXN[CCO-CYTOSOL]-FRUCTOSE-16-DIPHOSPHATE/WATER//FRUCTOSE-6P/Pi.59.'
# ]

Track atom flow in reactions

The atom mappings in SMILES format can be parsed with RDKit.

Reversibility mismatch in SBML and dat file

Here's the statistics for HumanCyc:

reactionDirection	reversible	COUNT(*)
"left_to_right"	false	871
"physiol_left_to_right"	false	1519
"physiol_right_to_left"	false	172
"right_to_left"	false	87
"reversible"	true	239
"reversible"	false	17
null	true	146
"physiol_left_to_right"	true	1

The reactionDirection field comes from reactions.dat, and reversible comes from the SBML file. To match with the results shown in HumanCyc:

• When reactionDirection is "reversible" and reversible is false, the reactions are reverisble.
• When reactionDirection is missing and reversible is true, the direction is unknown.
• When reactionDirection is not reversible and reversible is true, the reaction is irreversible.

In summary, the reversible property is not reliable and can be dropped.

reaction side doesn't match in SBML and dat file

For example, in PEPDEPHOS-RXN pyruvate is listed in the products (right) side, but in the reactions.dat file it is on the left-hand-side. This renders reactionDirection useless, because we don't really know which metabolites are on which side.

Additional information in reactions.dat

• Gibbs energy, reaction direction, systematic name, synonyms, etc.
• Associated pathways
• Citations

[MetaCyc] GIBBS-0 and STD-REDUCTION-POTENTIAL in reversible reactions

When parsing the reactions.dat file, we're matching the entries to the nodes in the graph based on the UNIQUE-ID and metaId. However, the sides in reactions.dat and metabolic-reactions.xml could be swapped, i.e. LEFT in the dat file could be Product in the SBML file.

For irreversible reactions, this is fixed in MetacycClient.fix_reaction_direction where we always assume the directions in the SBML file are true. For reversible reactions, this becomes a problem as GIBBS-0 and STD-REDUCTION-POTENTIAL could be reversed.

Pathway nodes with Reaction IDs

Some Pathway nodes are composite reactions and their info are stored in reactions.dat, e.g. RXN-4602.