hazyresearch / ddbiolib Goto Github PK

Build in some tooling to show the provenance of a supplied set of terms given some user configured set of ontologies/dictionaries.

Let the user specify an AUI path in a given ontology (source->target nodes) and then expand each node along the path using CUI mapping. Let user restrict by SAB and preferred_term flag

This is a weird package since it isn't available via pip. It's also not needed unless you directly use the Metathesarus object. Replace with more general solution.

G6PD A- | G6PD A -
Wilm's tumor

e.g., mysql-connector-python what version?
2.0.4

For dictionary building - given a UMLS ontology and a list of parents ignore all subtrees OR use only those subtrees for constructing dictionary entries.

Given an UMLS ontology with hierarchical structure, add a function that specifies a root and list of target subtree nodes to prune.

In order to move toward relationship extraction over mentions, we need a generic function to load pubtator annotations and map the character offsets from the title/abstract text to token offsets from corenlp. The output would be a Entities/Candidates object with all pubtator annotations from individual documents as individual entities.

Add doc_id to parsing step.

ddbiolib/examples/notebooks/* . file didn't find

score, partial matches, classification errors -- these should live outside the corpus object

I found a bug in match function, here is the example:
palmoplantar keratoderma (PPK) conditions | palmoplantar keratoderma -LRB- PPK -RRB- conditions
Left is the ground truth, right is our extracted mention.
It should match, since -LRB- equals to '(', and -RRB- equals to ')'.

No reason to extract all PMC archive files, for example, if we can just iterate over some compressed .zip/.bz2 files

If an ontology is in OWL/OBO format and not in UMLS, add support for loading and parsing that ontology to build dictionaries from it. Consider using rdflib for this.

Currently the UMLS module only supports concept (CUI) hierarchical structure for semantic types. This is often noisy and incorrect for any fine grained analysis and can break semantics when looking at hierarchical structure across different curated ontologies (SNOMED-CT, FMA, etc.).