AccFG: Accurate Functional Group Extraction and Molecular Structure Comparison

🚀News

• Try AccFG in Colab.

• (9/22/2025) AccFG v0.0.7: Fix bugs during import

• AccFG v0.0.4: Update FG names with lowercase; Update 3 new FGs (chloroformate etc.)

• AccFG v0.0.3:

  • update AccFG.run_mol() for directing processing rdkit Mol object
  • Lite version of AccFg is available through AccFG(lite=True), this will load a simplified FG list (e.g., no primary/secondary hydroxyl but only hydroxyl)

📝Introduction

This is the official code repository for the paper AccFG: Accurate Functional Group Extraction and Molecular Structure Comparison. AccFG is a tool for precise functional group (FG) extraction and molecular structure comparison.

📥Installation

We provide two methods to install AccFG:

Method 1: Installation by pip (recommended)

pip install accfg

Method 2: Installation from GitHub repository

To install AccFG, follow these steps:

1. Clone/download the repository and navigate to the project directory:

   git clone https://github.com/xuanliugit/AccFG.git
   cd AccFG

2. Install the required dependencies:

   conda create --name accfg python=3.10
   conda activate accfg
   pip install -r requirements.txt 
   # Or "pip install -e ."

Functional groups in AccFG

The FG dictionary is stored in ./accfg/fgs_common.csv and ./accfg/fgs_heterocycle.csv.

Note: The two fgs*.csv files are custom-formatted to be compatible with the AccFG.csv_to_dict() function in ./accfg/main.py. Compared to standard CSV files, they include additional annotation syntax to support structured parsing. Lines that begin with % are treated as comments to enhance readability and are excluded during data import.

⚙️Usage

Quick start:

# Get functional groups from SMILES
python run_accfg.py 'CN(C)/N=N/C1=C(NC=N1)C(=O)N'

# Compare two molecules
python run_accfg.py 'CNC(=O)Cc1nc(-c2ccccc2)cs1' --compare_smi 'CCNCCc1nc2ccccc2s1'

FG extraction

To extract functional groups:

# example.py
from accfg import AccFG
from accfg.draw import print_fg_tree

afg = AccFG(print_load_info=True)
smi = 'CN(C)/N=N/C1=C(NC=N1)C(=O)N'

fgs,fg_graph = afg.run(smi, show_atoms=True, show_graph=True)

print_fg_tree(fg_graph, fgs.keys(), show_atom_idx=True)
'''
├──Primary amide: ((10, 12, 11),)
...
'''
print(fgs)
'''
{'Primary amide': [(10, 12, 11)], 'Triazene': [(1, 3, 4)], 'imidazole': [(5, 9, 8, 7, 6)]}
'''

User-defined FGs Example:

# example.py
from accfg import AccFG
from accfg.draw import print_fg_tree

my_fgs_dict = {'Cephem': 'O=C(O)C1=CCS[C@@H]2CC(=O)N12', 'Thioguanine': 'Nc1nc(=S)c2[nH]cnc2[nH]1'}
my_afg = AccFG(user_defined_fgs=my_fgs_dict,print_load_info=True)

cephalosporin_C = 'CC(=O)OCC1=C(N2[C@@H]([C@@H](C2=O)NC(=O)CCC[C@H](C(=O)O)N)SC1)C(=O)O'
fgs,fg_graph = my_afg.run(cephalosporin_C, show_atoms=True, show_graph=True)

print_fg_tree(fg_graph, fgs.keys(), show_atom_idx=True) # This will print the FG tree

'''
├──Primary aliphatic amine: ((21,),)
├──...
'''

To print functional groups:

print(fgs) # Show top level FGs
'''
{'Primary aliphatic amine': [(21,)],
 'Carboxylic acid': [(22, 23, 24)],
 'Carboxylic ester': [(1, 2, 3, 4)],
 'Secondary amide': [(15, 16, 14, 13)],
 'Cephem': [(8, 7, 9, 6, 5, 27, 26, 25, 13, 11, 12, 10)]}
'''

FG extraction visualization

from accfg.draw import draw_mol_with_fgs, molimg

molimg(draw_mol_with_fgs(cephalosporin_C, afg=my_afg, img_size=(900,900)))

This will show image with FGs highlighted

Molecular structure comparison

from accfg import AccFG, compare_mols
from accfg.draw import draw_compare_mols, draw_RascalMCES, img_grid

smi_1,smi_2 = ('CNC(=O)Cc1nc(-c2ccccc2)cs1','CCNCCc1nc2ccccc2s1')
diff = compare_mols(smi_1, smi_2)
print(diff) # This print the structure difference
'''
(([('Secondary amide', 1, [(2, 3, 1)]),
   ...
'''

draw_RascalMCES(smi_1, smi_2) # This draw the RascalMCES comparison

Molecular structure comparison visualization

img = img_grid(draw_compare_mols(smi_1, smi_2),num_columns=2)
with open('results/compare_mols.png', 'wb') as f:
    img.save(f, format='PNG')
img

⚒️Run

To run the BBBP dataset, Lipophilicity dataset, BACE dataset, and CHEMBL drugs, simply run:

python run_data.py

The result is in ./molecule_data. The code to process the data is in exam_data.py

All other examples in the manuscript are in example.ipynb.

Cite this work

@article{liu2025accfg,
  title={AccFG: Accurate Functional Group Extraction and Molecular Structure Comparison},
  author={Liu, Xuan and Swaminathan, Sarathkrishna and Zubarev, Dmitry and Ransom, Brandi and Park, Nathaniel and Schmidt, Kristin and Zhao, Huimin},
  journal={Journal of Chemical Information and Modeling},
  volume={65},
  number={16},
  pages={8593--8602},
  year={2025},
  publisher={ACS Publications}
}