Image2InChI 2.0.3

Convert a picture to InChI

Image2InChI

descirption

This is the repository for Image2InChI, an image-to-graph model that translates a molecular image to its InChI.

Quick Start

Installation

Install Image2InChI with pip

pip install Image2InChI

Example

Predict based on a picture

from image2inchi.model import Image2InChI

model = Image2InChI(chk_pt='/weight/chk_pt.pth')

inchi_pred = model.pred_one('/data/aaa.png')
print(inchi_pred)

Using the pred_batch(), pass in a folder path to predict This function runs the Image2InChI model on all images in a given directory and saves the results in results.csv in the same directory. Each line of the output file contains the image file name and the InChI representation of the described molecule.

from image2inchi.model import Image2InChI

model = Image2InChI(chk_pt='/weight/chk_pt.pth')

inchi_df = model.pred_batch('/data/bms_test')
print(inchi_df)

Public Dataset

Validate using a publicly available dataset

from image2inchi.model import Image2InChI, test_bms1000, test_jpo, test_clef, test_uob, test_uspto

model = Image2InChI(chk_pt='/weight/chk_pt.pth')

test_bms1000(model)
test_jpo(model)
test_clef(model)
test_uob(model)
test_uspto(model)

You can use the test_public_dataset() to verify the above

from image2inchi.model import Image2InChI, test_public_dataset

model = Image2InChI(chk_pt='/weight/chk_pt.pth')
test_public_dataset(model)

Evaluation

scoring metrics can be obtained using the Scorer.scoring(). The function needs to pass in two iterable parameters, parameter 1 is the InChI label and parameter 2 is the predicted InChI.

from image2inchi.utils import Scorer

inchi_list = []  # InChI label list
inchi_pred_list = []  # predicted InChI list
inchi_acc, morgan_fp, mcs, lcs = Scorer.scoring(inchi_list, inchi_pred_list)
print(f'result: inchi_acc - {inchi_acc}, morgan_fp - {morgan_fp}, mcs - {mcs}, lcs - {lcs}')