Molecule Standardization and Filtering Pipeline

Standardize a molecule CSV dataset, remove PAINS compounds, filter by heavy atom count, and export a clean cheminformatics-ready dataset.

The main workflow is implemented in molecule_standardization.ipynb.

What the Notebook Does

1. Loads molecules from a CSV file.
2. Parses SMILES strings with RDKit.
3. Applies a standardization pipeline:
   • sanitize molecule
   • remove explicit hydrogens
   • disconnect metals
   • normalize functional groups
   • keep the largest organic fragment
   • reionize
   • neutralize charges
   • canonicalize tautomer
4. Counts PAINS alerts.
5. Filters molecules by:
   • successful parsing and standardization
   • heavy atom count greater than 10
   • zero PAINS alerts
6. Exports a clean CSV and optional SDF.
7. Saves plots and molecule preview images.

Requirements

• Python 3.8+
• Jupyter Notebook or JupyterLab
• pandas
• numpy
• tqdm
• matplotlib
• RDKit

Recommended installation with conda:

conda install -c conda-forge rdkit pandas numpy tqdm matplotlib jupyterlab

Input Data

The notebook expects a CSV file with a SMILES column.

The configuration cell uses these defaults:

INPUT_CSV = "reframe_raw.csv"
SMILES_COL = "smiles"
MOL_ID_COL = "source_id"
OUTPUT_CSV = "skc_outputs/reframe_std.csv"

Update those values before running the notebook.

Example input columns:

smiles,name,source_id
CCO,ethanol,CMPD001
c1ccccc1,benzene,CMPD002

Usage

Start Jupyter:

jupyter lab

Open:

molecule_standardization.ipynb

Then run the notebook from top to bottom after editing the configuration cell.

Outputs

By default, the notebook writes files under skc_outputs/.

Expected outputs include:

• reframe_std.csv: standardized and filtered molecule table
• reframe_std.sdf: optional SDF export
• reframe_distribution_plots.png: PAINS and heavy atom count plots
• preview_reframe_mols.png: grid preview of filtered molecules

The CSV includes key columns such as:

• std_smiles
• heavy_atom_count
• pains_alert_count
• pains_alert_names
• original input metadata columns

Filtering Rules

The default filtering logic keeps molecules where:

• std_ok == True
• heavy_atom_count > 10
• pains_alert_count == 0

Change these thresholds in the configuration cell:

MIN_HEAVY_ATOMS = 10
MAX_PAINS_ALERTS = 0

Notes

This notebook is designed as a preprocessing step before downstream cheminformatics tasks such as diversity selection, clustering, docking preparation, or QSAR modeling.