pumpkin-py 0.0.2

PumpkinPy - Semantic similarity implemented in python

About

PumpkinPy uses IC ordered bitmaps for fast ranking of genes and diseases (phenotypes are sorted by descending frequency and one-hot encoded). This is useful for larger ontologies such as Upheno and large datasets such as ranking all mouse genes given a set of input HPO terms. This approach was first used in OWLTools and OwlSim-v3.

The goal of this project was to build an implementation of the PhenoDigm algorithm in python. There are also implementations for common measures for distance and similarity (euclidean, cosine, Jin-Conrath, Resnik, jaccard)

Disclaimer: This is a side project needs more documentation and testing

Getting Started

Requires python 3.8+ and python3-dev to install pyroaring

Installing from pypi

pip install pumpkin_py

Building locally

To build locally first install poetry -

https://python-poetry.org/docs/#installation

Then run make:

make

Usage

Get a list of implemented similarity measures

from pumpkin_py import get_methods
get_methods()
['jaccard', 'cosine', 'phenodigm', 'symmetric_phenodigm', 'resnik', 'symmetric_resnik', 'ic_cosine', 'sim_gic']

Load closures and annotations

import gzip
from pathlib import Path

from pumpkin_py import build_ic_graph_from_closures, flat_to_annotations, search

closures = Path('.') / 'data' / 'hpo' / 'hp-closures.tsv.gz'
annotations = Path('.') / 'data' / 'hpo' / 'phenotype-annotations.tsv.gz'

root = "HP:0000118"

with gzip.open(annotations, 'rt') as annot_file:
    annot_map = flat_to_annotations(annot_file)

with gzip.open(closures, 'rt') as closure_file:
    graph = build_ic_graph_from_closures(closure_file, root, annot_map)

Search for the best matching disease given a phenotype profile

import pprint
from pumpkin_py import search

profile_a = (
    "HP:0000403,HP:0000518,HP:0000565,HP:0000767,"
    "HP:0000872,HP:0001257,HP:0001263,HP:0001290,"
    "HP:0001629,HP:0002019,HP:0002072".split(',')
)

search_results = search(profile_a, annot_map, graph, 'phenodigm')

pprint.pprint(search_results.results[0:5])

[SimMatch(id='ORPHA:94125', rank=1, score=72.67599348696685),
 SimMatch(id='ORPHA:79137', rank=2, score=71.57368233248252),
 SimMatch(id='OMIM:619352', rank=3, score=70.98305459477629),
 SimMatch(id='OMIM:618624', rank=4, score=70.94596234638497),
 SimMatch(id='OMIM:617106', rank=5, score=70.83097366257857)]

Example scripts for fetching Monarch annotations and closures

Uses robot and sparql to generate closures and class labels

Annotation data is fetched from the latest Monarch release

• Requires >Java 8

cd data/monarch/ && make

PhenoDigm Reference: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3649640/
Exomiser: https://github.com/exomiser/Exomiser
OWLTools: https://github.com/owlcollab/owltools
OWLSim-v3: https://github.com/monarch-initiative/owlsim-v3