Molecular Descriptors - Usage Guide

Overview

Calculate molecular fingerprints and physicochemical properties for compound characterization, drug-likeness assessment, and machine learning featurization.

Prerequisites

pip install rdkit
pip install numpy pandas

Quick Start

Tell your AI agent what you want to do: - "Calculate ECFP4 fingerprints for my compound library" - "Check Lipinski rule of 5 compliance for these molecules" - "Calculate QED drug-likeness scores" - "Generate descriptors for machine learning"

Example Prompts

Fingerprints

"Generate Morgan fingerprints with radius 2 and 2048 bits for my molecules."

"Calculate MACCS keys for similarity searching."

Drug-Likeness

"Check which compounds pass Lipinski's rule of 5."

"Calculate QED scores and apply QED > 0.5 as a documented project triage heuristic; show the unfiltered distribution too."

Full Descriptor Set

"Calculate all available RDKit descriptors for my molecules."

"Generate 3D conformers and calculate shape descriptors."

What the Agent Will Do

Load molecules from provided structures
Calculate requested descriptors/fingerprints
Compile results into a DataFrame
Apply filters based on thresholds if requested
Export results in requested format

Tips

ECFP4 = radius 2 and ECFP6 = radius 3 (the ECFP number denotes diameter, 2 * radius)
Include useChirality=True for stereo-sensitive fingerprints
QED > 0.5 can be used as a repository/project triage heuristic, but calibrate it for the library and do not treat it as a universal drug-likeness boundary
RDKit defaults to ETKDG; select AllChem.ETKDGv3() explicitly when the v3 small-ring and macrocycle improvements are intended
Lipinski thresholds: MW <= 500, LogP <= 5, HBD <= 5, HBA <= 10
3D descriptors require conformer generation first

molecular-io - Load molecules for descriptor calculation
similarity-searching - Use fingerprints for similarity
admet-prediction - Predict ADMET from descriptors
machine-learning/biomarker-discovery - ML on molecular features