preon (PREcision Oncology Normalization)

preon is a fuzzy search tool for medical entities.

Installation

You can install preon with PyPi: python -m pip install preon

Examples

Let's first import the normalizer and EBI drug names with CHEMBL ids.

>>> from preon.normalization import PrecisionOncologyNormalizer
>>> from preon.drug import store_ebi_drugs, load_ebi_drugs

Please download the EBI compound CSV file and store it as a local resource. This step only has to be performed when the resource file is created or updated.

>>> store_ebi_drugs("/Users/Username/Downloads/compounds.csv")

Next, we can fit the normalizer with the drug names and ids as its reference data.

>>> drug_names, chembl_ids = load_ebi_drugs()
>>> normalizer = PrecisionOncologyNormalizer().fit(drug_names, chembl_ids)

We can now search for drug names and retrieve their CHEMBL ids. Let's search for the cancer drug "Avastin".

>>> normalizer.query("Avastin")
(['avastin'], [['CHEMBL1201583']], {'match_type': 'exact'})

As a result for our query, we get list of matching normalized drug names (in this case ['avastin']), a list of associated CHEMBL ids for every returned drug name [['CHEMBL1201583']] and some meta information about the matching {'match_type': 'exact'}. We can also search for multi-token drug names like "Ixabepilone Epothilone B analog" and find CHEMBL ids for the relevant tokens.

>>> normalizer.query("Ixabepilone Epothilone B analog")
(['ixabepilone'], [['CHEMBL1201752']], {'match_type': 'substring'})

We find the relevant drug name ['ixabepilone'] and preon provides the meta information that the matching is based on a substring. On default, preon only looks for 1 matching token. It can also look for n-grams by setting the n_grams parameter in the query method. Let's take a harder example, say "Isavuconazonium", but misspell it as "Isavuconaconium".

>>> normalizer.query("Isavuconaconium")
(['isavuconazonium'], [['CHEMBL1183349']], {'match_type': 'partial', 'edit_distance': 0.067})

preon finds the correct drug "Isavuconazonium" and provides the meta information that it is a partial match with 7% distance. It returns drug names with a distance smaller than 20% on default. In order to change this parameter, set the threshold argument in the query method. If preon cannot normalize the query, it returns None and issues a user warning.

>>> normalizer.query("risolipase en.")
preon/normalization.py:50: UserWarning: Cannot match risolipase en. to reference data. Try changing the partial matching threshold or number of n-grams.

For automatic data integrations, warnings can be stored in a logging file, see e.g. here. In a similar fashion, you can also normalize cancer types or genes. We provide gold standards for preon with which we test it. For more detail, see the example notebooks. We also use preon in practice to normalize and integrate medical data in the PREDICT project.

Citation

The preon package is actively maintained, updated and intended for application. If you use it in your scientific publication, we would appreciate the following citation:

@article {preon2023,
	author = {Arik Ermshaus and Michael Piechotta and Gina R{\"u}ter and Ulrich Keilholz and Ulf Leser and Manuela Benary},
	title = {preon: Fast and accurate entity normalization for drug names and cancer types in precision oncology},
	year = {2023},
	doi = {10.1101/2023.05.22.540912},
	journal = {bioRxiv}
}