Prediction Functions Demo

This Jupyter notebook demonstrates the usage of the prediction functions that are part of ChemREL.

Tip

To run a copy of this notebook yourself, download the corresponding predict_demo.ipynb file here.

Import ChemREL

To begin, import the predict submodule from the functions package in ChemREL, as follows.

from chemrel.functions import predict

Predict Submodule

The predict submodule contains two available functions, predict_span() and predict_rel(). Their usage is demonstrated below. For the full documentation of each function, see the ChemREL Functions Reference.

We first define the text string to predict as a variable, example_text.

example_text = '''Two physical properties of the fluorinated carboxylic acid,
2H-hexadecafluoro-2-decenoic acid (C10F16H2O2), were investigated in this study: 
melting point and solubility. Melting point data were ascertained on a traditional melting point
apparatus and a differential scanning calorimeter (DSC), while solubility data were obtained using an LC−MS/MS system.
A melting point of 105 ± 1 °C and an aqueous solubility of 64 ± 5 ng/μL at ambient temperatures were observed'''

Span Categorization

We demonstrate how predict_span() can be used to predict spans, as follows. This assumes that the span categorization model provided by ChemREL has been initialized at chemrelmodels/sc/tok2vec.

spans_dict = predict.predict_span("chemrelmodels/sc/tok2vec", example_text)

We can now print the contents of spans_dict to view the predicted spans.

print(spans_dict)

{'CHEMICAL': [('C10F16H2O2', 0.9998276), ('fluorinated carboxylic acid', 0.99959916), ('2H-hexadecafluoro-2-decenoic acid', 0.9974232)], 'MP': [('105 ± 1 °C', 0.99998033)]}

Relation Prediction

We demonstrate how predict_rel() can be used to predict relations, as follows. This assumes that the span categorization and relation extraction models provided by ChemREL have been initialized at chemrelmodels/sc/tok2vec and chemrelmodels/rel/tok2vec, respectively.

rels_dict = predict.predict_rel("chemrelmodels/sc/tok2vec", "chemrelmodels/rel/tok2vec", example_text)

We can now print the contents of rels_dict to view the predicted relations.

print(rels_dict)

{('CHEMICAL', 'MP'): [(('fluorinated carboxylic acid', '105 ± 1 °C'), 0.999992), (('2H-hexadecafluoro-2-decenoic acid', '105 ± 1 °C'), 0.9999999), (('C10F16H2O2', '105 ± 1 °C'), 0.9999999)]}