Home

Awesome

<p align="center"><img width="50%" src="docs/scispacy-logo.png" /></p>

This repository contains custom pipes and models related to using spaCy for scientific documents.

In particular, there is a custom tokenizer that adds tokenization rules on top of spaCy's rule-based tokenizer, a POS tagger and syntactic parser trained on biomedical data and an entity span detection model. Separately, there are also NER models for more specific tasks.

Just looking to test out the models on your data? Check out our demo (Note: this demo is running an older version of scispaCy and may produce different results than the latest version).

Installation

Installing scispacy requires two steps: installing the library and intalling the models. To install the library, run:

pip install scispacy

to install a model (see our full selection of available models below), run a command like the following:

pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.5.4/en_core_sci_sm-0.5.4.tar.gz

Note: We strongly recommend that you use an isolated Python environment (such as virtualenv or conda) to install scispacy. Take a look below in the "Setting up a virtual environment" section if you need some help with this. Additionally, scispacy uses modern features of Python and as such is only available for Python 3.6 or greater.

Installation note: nmslib

Over the years, installing nmslib has becomes quite difficult. There are a number of GitHub issues on scispaCy and the nmslib repo itself about this. This matrix is an attempt to help users install nmslib in whatever environment they have. I don't have access to every type of environment, so if you are able to test something out, please open an issue or pull request!

Windows 11Windows Subsystem for LinuxMac M1Mac M2Mac M3Mac M4Intel Mac
Python 3.8✅✅💻❓❓❌🔨❓
Python 3.9❌🐍✅💻❌🔨❓❌🔨❓
Python 3.10❌🐍✅❓❓❓❌🔨✅
Python 3.11❌🐍❌🐍❓❓❓❌🔨❌
Python 3.12❌🐍❌🐍🧠❓❓❓❌🧠❓

✅ = works normally with pip install of scispacy

❌ = does not work normally with pip install of scispacy

🐍 = can be installed with mamba install nmslib

💻 = can be installed with CFLAGS="-mavx -DWARN(a)=(a)" pip install nmslib

🧠 = can be installed with pip install nmslib-metabrainz

🔨 = can be installed with conda install -c conda-forge nmslib

❓ = unconfirmed

Other methods mentioned in GitHub issues, but unconfirmed what versions they work for:

Setting up a virtual environment

Mamba can be used set up a virtual environment with the version of Python required for scispaCy. If you already have a Python environment you want to use, you can skip to the 'installing via pip' section.

  1. Follow the installation instructions for Mamba.

  2. Create a Conda environment called "scispacy" with Python 3.9 (any version >= 3.6 should work):

    mamba create -n scispacy python=3.10

  3. Activate the Mamba environment. You will need to activate the Conda environment in each terminal in which you want to use scispaCy.

    mamba activate scispacy

Now you can install scispacy and one of the models using the steps above.

Once you have completed the above steps and downloaded one of the models below, you can load a scispaCy model as you would any other spaCy model. For example:

import spacy
nlp = spacy.load("en_core_sci_sm")
doc = nlp("Alterations in the hypocretin receptor 2 and preprohypocretin genes produce narcolepsy in some animals.")

Note on upgrading

If you are upgrading scispacy, you will need to download the models again, to get the model versions compatible with the version of scispacy that you have. The link to the model that you download should contain the version number of scispacy that you have.

Available Models

To install a model, click on the link below to download the model, and then run

pip install </path/to/download>

Alternatively, you can install directly from the URL by right-clicking on the link, selecting "Copy Link Address" and running

pip install CMD-V(to paste the copied URL)
ModelDescriptionInstall URL
en_core_sci_smA full spaCy pipeline for biomedical data with a ~100k vocabulary.Download
en_core_sci_mdA full spaCy pipeline for biomedical data with a ~360k vocabulary and 50k word vectors.Download
en_core_sci_lgA full spaCy pipeline for biomedical data with a ~785k vocabulary and 600k word vectors.Download
en_core_sci_scibertA full spaCy pipeline for biomedical data with a ~785k vocabulary and allenai/scibert-base as the transformer model. You may want to use a GPU with this model.Download
en_ner_craft_mdA spaCy NER model trained on the CRAFT corpus.Download
en_ner_jnlpba_mdA spaCy NER model trained on the JNLPBA corpus.Download
en_ner_bc5cdr_mdA spaCy NER model trained on the BC5CDR corpus.Download
en_ner_bionlp13cg_mdA spaCy NER model trained on the BIONLP13CG corpus.Download

Additional Pipeline Components

AbbreviationDetector

The AbbreviationDetector is a Spacy component which implements the abbreviation detection algorithm in "A simple algorithm for identifying abbreviation definitions in biomedical text.", (Schwartz & Hearst, 2003).

You can access the list of abbreviations via the doc._.abbreviations attribute and for a given abbreviation, you can access it's long form (which is a spacy.tokens.Span) using span._.long_form, which will point to another span in the document.

Example Usage

import spacy

from scispacy.abbreviation import AbbreviationDetector

nlp = spacy.load("en_core_sci_sm")

# Add the abbreviation pipe to the spacy pipeline.
nlp.add_pipe("abbreviation_detector")

doc = nlp("Spinal and bulbar muscular atrophy (SBMA) is an \
           inherited motor neuron disease caused by the expansion \
           of a polyglutamine tract within the androgen receptor (AR). \
           SBMA can be caused by this easily.")

print("Abbreviation", "\t", "Definition")
for abrv in doc._.abbreviations:
	print(f"{abrv} \t ({abrv.start}, {abrv.end}) {abrv._.long_form}")

>>> Abbreviation	 Span	    Definition
>>> SBMA 		 (33, 34)   Spinal and bulbar muscular atrophy
>>> SBMA 	   	 (6, 7)     Spinal and bulbar muscular atrophy
>>> AR   		 (29, 30)   androgen receptor

Note If you want to be able to serialize your doc objects, load the abbreviation detector with make_serializable=True, e.g. nlp.add_pipe("abbreviation_detector", config={"make_serializable": True})

EntityLinker

The EntityLinker is a SpaCy component which performs linking to a knowledge base. The linker simply performs a string overlap - based search (char-3grams) on named entities, comparing them with the concepts in a knowledge base using an approximate nearest neighbours search.

Currently (v2.5.0), there are 5 supported linkers:

You may want to play around with some of the parameters below to adapt to your use case (higher precision, higher recall etc).

This class sets the ._.kb_ents attribute on spacy Spans, which consists of a List[Tuple[str, float]] corresponding to the KB concept_id and the associated score for a list of max_entities_per_mention number of entities.

You can look up more information for a given id using the kb attribute of this class:

print(linker.kb.cui_to_entity[concept_id])

Example Usage

import spacy
import scispacy

from scispacy.linking import EntityLinker

nlp = spacy.load("en_core_sci_sm")

# This line takes a while, because we have to download ~1GB of data
# and load a large JSON file (the knowledge base). Be patient!
# Thankfully it should be faster after the first time you use it, because
# the downloads are cached.
# NOTE: The resolve_abbreviations parameter is optional, and requires that
# the AbbreviationDetector pipe has already been added to the pipeline. Adding
# the AbbreviationDetector pipe and setting resolve_abbreviations to True means
# that linking will only be performed on the long form of abbreviations.
nlp.add_pipe("scispacy_linker", config={"resolve_abbreviations": True, "linker_name": "umls"})

doc = nlp("Spinal and bulbar muscular atrophy (SBMA) is an \
           inherited motor neuron disease caused by the expansion \
           of a polyglutamine tract within the androgen receptor (AR). \
           SBMA can be caused by this easily.")

# Let's look at a random entity!
entity = doc.ents[1]

print("Name: ", entity)
>>> Name: bulbar muscular atrophy

# Each entity is linked to UMLS with a score
# (currently just char-3gram matching).
linker = nlp.get_pipe("scispacy_linker")
for umls_ent in entity._.kb_ents:
	print(linker.kb.cui_to_entity[umls_ent[0]])


>>> CUI: C1839259, Name: Bulbo-Spinal Atrophy, X-Linked
>>> Definition: An X-linked recessive form of spinal muscular atrophy. It is due to a mutation of the
  				gene encoding the ANDROGEN RECEPTOR.
>>> TUI(s): T047
>>> Aliases (abbreviated, total: 50):
         Bulbo-Spinal Atrophy, X-Linked, Bulbo-Spinal Atrophy, X-Linked, ....

>>> CUI: C0541794, Name: Skeletal muscle atrophy
>>> Definition: A process, occurring in skeletal muscle, that is characterized by a decrease in protein content,
                fiber diameter, force production and fatigue resistance in response to ...
>>> TUI(s): T046
>>> Aliases: (total: 9):
         Skeletal muscle atrophy, ATROPHY SKELETAL MUSCLE, skeletal muscle atrophy, ....

>>> CUI: C1447749, Name: AR protein, human
>>> Definition: Androgen receptor (919 aa, ~99 kDa) is encoded by the human AR gene.
                This protein plays a role in the modulation of steroid-dependent gene transcription.
>>> TUI(s): T116, T192
>>> Aliases (abbreviated, total: 16):
         AR protein, human, Androgen Receptor, Dihydrotestosterone Receptor, AR, DHTR, NR3C4, ...

Hearst Patterns (v0.3.0 and up)

This component implements Automatic Aquisition of Hyponyms from Large Text Corpora using the SpaCy Matcher component.

Passing extended=True to the HyponymDetector will use the extended set of hearst patterns, which include higher recall but lower precision hyponymy relations (e.g X compared to Y, X similar to Y, etc).

This component produces a doc level attribute on the spacy doc: doc._.hearst_patterns, which is a list containing tuples of extracted hyponym pairs. The tuples contain:

Usage:

import spacy
from scispacy.hyponym_detector import HyponymDetector

nlp = spacy.load("en_core_sci_sm")
nlp.add_pipe("hyponym_detector", last=True, config={"extended": False})

doc = nlp("Keystone plant species such as fig trees are good for the soil.")

print(doc._.hearst_patterns)
>>> [('such_as', Keystone plant species, fig trees)]

Citing

If you use ScispaCy in your research, please cite ScispaCy: Fast and Robust Models for Biomedical Natural Language Processing. Additionally, please indicate which version and model of ScispaCy you used so that your research can be reproduced.

@inproceedings{neumann-etal-2019-scispacy,
    title = "{S}cispa{C}y: {F}ast and {R}obust {M}odels for {B}iomedical {N}atural {L}anguage {P}rocessing",
    author = "Neumann, Mark  and
      King, Daniel  and
      Beltagy, Iz  and
      Ammar, Waleed",
    booktitle = "Proceedings of the 18th BioNLP Workshop and Shared Task",
    month = aug,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/W19-5034",
    doi = "10.18653/v1/W19-5034",
    pages = "319--327",
    eprint = {arXiv:1902.07669},
    abstract = "Despite recent advances in natural language processing, many statistical models for processing text perform extremely poorly under domain shift. Processing biomedical and clinical text is a critically important application area of natural language processing, for which there are few robust, practical, publicly available models. This paper describes scispaCy, a new Python library and models for practical biomedical/scientific text processing, which heavily leverages the spaCy library. We detail the performance of two packages of models released in scispaCy and demonstrate their robustness on several tasks and datasets. Models and code are available at https://allenai.github.io/scispacy/.",
}

ScispaCy is an open-source project developed by the Allen Institute for Artificial Intelligence (AI2). AI2 is a non-profit institute with the mission to contribute to humanity through high-impact AI research and engineering.