Conformer Generation Usage Guide

Overview

Generate 3D conformer ensembles for molecules from 2D structures. Default to ETKDGv3 (Wang et al. 2020, building on Riniker & Landrum 2015) with MMFF94 optimization for drug-like molecules, and use CREST with GFN2-xTB for difficult macrocycles or peptides. Treat GeoMol as a research alternative whose released-model coverage and target chemistry require validation.

Prerequisites

pip install rdkit
# For CREST + GFN2-xTB (semi-empirical):
conda install -c conda-forge xtb crest

Quick Start

Tell the AI agent what to do: - "Generate 50 ETKDGv3 conformers for this SMILES with MMFF94 optimization" - "Build a 3D conformer ensemble suitable for docking input" - "Sample macrocycle conformers using CREST + GFN2-xTB" - "Generate Boltzmann-weighted descriptor averages over a 100-conformer ensemble" - "Prune conformers below 0.5 RMSD and within a 5 kcal/mol energy window"

Example Prompts

Drug-like conformer ensemble

"Generate 20 ETKDGv3 conformers for each compound in library.csv. MMFF94 optimize. Prune at RMSD < 0.5 A and energy window 10 kcal/mol. Output SDF with conformer index in property."

Macrocycle high-quality sampling

"Sample 200 conformers of cyclosporine (CAS 59865-13-3) with CREST + GFN2-xTB. Filter to 5 kcal/mol energy window. Output cluster centroids."

3D QSAR descriptor input

"For each compound in active_set.sdf, generate 50 ETKDGv3 conformers and compute Boltzmann-averaged 3D descriptors (asphericity, eccentricity, PMI). Output descriptor table."

Docking input

"Generate a single low-energy ETKDGv3 + MMFF94 conformer for each SMILES; write to multi-SDF for downstream Vina docking."

What the Agent Will Do

1. Parse SMILES, add hydrogens, embed initial 3D coordinates with ETKDGv3.
2. Generate the requested number of conformers (using EmbedMultipleConfs).
3. Optimize each conformer with MMFF94s; fall back to UFF only after checking parameter coverage, and record convergence failures.
4. Compute energy of each conformer.
5. Prune by RMSD and energy window.
6. For macrocycles or complex molecules: optionally pipe to CREST + GFN2-xTB.
7. Output: SDF / XYZ with conformer ensemble + energies + properties.

Tips

• As a starting repository heuristic, use n_conf = max(10, 5 * NumRotatableBonds + 10), then increase sampling until the downstream metric is stable.
• For docking and 3D descriptors, increase the ensemble until the downstream result is stable; fixed counts are only starting budgets.
• Do not interpret raw MMFF/UFF energies as thermodynamic populations without validating the population model.
• For macrocycles: use useMacrocycleTorsions=True; benchmark difficult cases with the higher-cost CREST workflow.
• Set randomSeed for reproducibility.
• Always AddHs() before embedding.

Related Skills

• chemoinformatics/molecular-io - Parse molecules
• chemoinformatics/molecular-descriptors - 3D descriptors from ensembles
• chemoinformatics/shape-similarity - Multi-conformer 3D shape searching
• chemoinformatics/virtual-screening - Generate 3D for docking
• chemoinformatics/free-energy-calculations - Conformers for MD/FEP setup
• chemoinformatics/pharmacophore-modeling - 3D pharmacophore from ensembles