humumls

UMLS in Python with MongoDB.

This package allows for the conversion of UMLS .RRF files to a document-level MongoDB database. Note that the goal of this package is not to provide a 1-to-1 mapping from the UMLS .RRF mapping to a MongoDB, but rather to provide useful abstractions where necessary.

Requirements

langid
pymongo
tqdm

See requirements.txt for specific package information

Important changes

• The Atom Unique Identifier (AUI) has been removed, as this is no longer necessary when using document-level databases.
• There is no distinction between different sources, as the SAB field has been removed.
• All strings with a STR field over 1000 bytes in length are considered noisy, and removed.
• Opaque field names, such as relations, have been expanded to their natural language alternatives.
• Definitions are tagged with a language using langid.
• All information not in MRDEF, MRCONSO or MRREL is discarded.

Usage

First, make sure that you have a MongoDB instance running, and that you know on which port and hostname it is running.

The tablecreator.py file contains the code for creating the MongoDB.

from humumls import createdb

# Get all English and Dutch items.
languages = ["ENG", "DUT"]

# Use default port, host and name.
# Create the tables. This takes ~2 hours on a normal laptop.
createdb(languages, "path/to/meta")

This MongoDB can be addressed using pymongo, or the provided aggregate class. The aggregate class is meant to be expanded for your specific purposes. Currently, it contains examples of use.

from examples.aggregator import Aggregator

# You can define your own aggregate queries.
agg = Aggregator()

# Query using connections.
concept = agg.concept["C0032344"]
string = agg.string["S000124"]

# Perform aggregate queries.
cancers = agg.concepts_string("cancer")