---
name: conformer-ensemble-energy-ranking
description: Use when after RDKit has generated a large set of 3D conformers for a molecule in SDF or XYZ format, and before submitting conformers to computationally expensive quantum-chemical methods (e.g., QUICK).
license: CC-BY-4.0
metadata:
  edam_operation: http://edamontology.org/operation_3938
  edam_topics:
  - http://edamontology.org/topic_0154
  - http://edamontology.org/topic_3314
  tools:
  - ASE-ANI
  - RDKit
  - QUICK
  - Snakemake
  techniques:
  - ion-mobility-MS
derived_from:
- doi: 10.1021/jasms.1c00315
  title: POMICS
evidence_spans:
- 'ASE-ANI: For conformation filtering'
claims: []
provenance:
  collection: https://w3id.org/holobiomicslab/asb-skill/collection/metabolomics/v2
  assembled_by: scripts/collect_metabolomics_collection.py
  sources:
  - build: coll_pomics_cq
    doi: 10.1021/jasms.1c00315
    title: POMICS
  dedup_kept_from: coll_pomics_cq
schema_version: 0.2.0
attribution:
  generator: AgenticScienceBuilder
  original_doi: 10.1021/jasms.1c00315
  all_source_dois:
  - 10.1021/jasms.1c00315
  zenodo_doi: 10.5281/zenodo.20794027
  curators: []
  promoter: Louis-Félix Nothias
  sponsor: CNRS & Université Côte d'Azur
---

# conformer-ensemble-energy-ranking

## Summary

Rank and filter conformer ensembles generated by RDKit using neural-network potential energies computed by ASE-ANI, retaining only the lowest-energy subset for downstream quantum-chemical calculations. This step reduces computational cost and focuses conformational sampling on thermodynamically favorable structures.

## When to use

After RDKit has generated a large set of 3D conformers for a molecule in SDF or XYZ format, and before submitting conformers to computationally expensive quantum-chemical methods (e.g., QUICK). Use this skill when you need to pre-screen conformers to avoid wasting HPC resources on high-energy, non-viable geometries.

## When NOT to use

- Input conformers are already filtered by a more rigorous method (e.g., molecular dynamics or explicit solvent sampling); re-filtering adds no value.
- The molecule contains elements outside the CHNO set supported by ANI-1x/ANI-1ccx (e.g., halogens, transition metals); ASE-ANI will fail or produce undefined predictions.
- Computational budget or HPC access permits direct quantum-chemical screening of all conformers; energy pre-filtering is unnecessary overhead.

## Inputs

- Conformer ensemble in SDF or XYZ format (output from RDKit conformation generation)
- ASE-ANI potential (ANI-1x or ANI-1ccx neural network model)
- Selection parameters: number of conformers to retain (N) or energy threshold (ΔE in kcal/mol)

## Outputs

- Filtered conformer ensemble (XYZ or compatible format)
- Energy-ranked conformer list (mapping conformer ID to computed energy in eV)
- Conformer subset metadata (indices, energies, selection criterion applied)

## How to apply

Load all RDKit-generated conformer structures (typically in SDF or XYZ format) into ASE-ANI. For each conformer, compute a single-point energy using the ANI-1x or ANI-1ccx neural-network potential, which provides DFT-quality predictions at force-field speed. Rank all conformers by their computed energy and apply one of two selection criteria: (1) retain the top N lowest-energy conformers (e.g., top 5–10), or (2) apply an energy threshold (e.g., retain all conformers within 5–10 kcal/mol of the global minimum). Export the filtered subset in a format compatible with downstream quantum-chemical software (e.g., XYZ for QUICK). The rationale is that ASE-ANI rapidly eliminates implausible conformations while preserving thermodynamically accessible states for Boltzmann-weighted CCS ensemble averaging.

## Related tools

- **RDKit** (Generates initial conformer ensemble; conformer-ensemble-energy-ranking filters its output) — https://www.rdkit.org
- **ASE-ANI** (Computes single-point neural-network potential energies for each conformer and enables ranking) — https://github.com/isayev/ASE_ANI
- **QUICK** (Downstream quantum-chemical tool that receives the filtered conformer subset for CCS calculation) — https://github.com/merzlab/QUICK
- **Snakemake** (Orchestrates the conformer generation, filtering, and quantum-calculation workflow with parallelization on HPC systems) — https://github.com/DasSusanta/snakemake_ccs

## Examples

```
python -c "from ase import Atoms; from ase_ani import ANI; import ase.io; structures = ase.io.read('conformers.sdf', index=':'); calc = ANI(); energies = []; [energies.append(s.calc_potential_energy(calc)) for s in structures]; top_n = sorted(zip(energies, range(len(energies))), key=lambda x: x[0])[:5]; [ase.io.write(f'filtered_{i}.xyz', structures[idx]) for e, idx in top_n]"
```

## Evaluation signals

- Energy values are computed and differ across the conformer ensemble (no single energy for all conformers; indicates ASE-ANI ran successfully).
- Filtered conformer count is either exactly N (if N-conformer selection) or all conformers fall within the specified energy threshold ΔE of the minimum.
- Output geometry files are valid XYZ or SDF with no structural corruption (parseable by downstream tools).
- Ensemble-averaged CCS values from the filtered conformers are stable and reproducible across re-filtering runs (convergence test).
- Energy ranking is monotonically increasing or flat; no out-of-order conformers in the ranked list.

## Limitations

- ASE-ANI is deprecated and no longer actively supported; users should transition to TorchANI (PyTorch implementation). The original ASE-ANI codebase requires Python 3.6, CUDA 9.2, and modern NVIDIA GPUs (compute capability ≥5.0) and works only on Ubuntu Linux variants.
- ASE-ANI potentials (ANI-1x, ANI-1ccx) support only CHNO elements; molecules containing halogens, phosphorus, sulfur, or transition metals will fail or produce unreliable predictions.
- Neural-network energies are approximate and trained on gas-phase geometries; they do not account for solvation, counterions, or pH effects relevant to experimental ESI-MS conditions. Post-filtering by quantum methods is essential for accuracy.
- Selection threshold (N or ΔE) is user-defined and influences downstream CCS statistics; no data-driven guidance is provided for optimal threshold choice. Over-filtering (small N) may miss conformers important for Boltzmann averaging; under-filtering (large N or ΔE) negates computational savings.

## Evidence

- [other] Load conformer structures generated by RDKit in a standard molecular format (e.g., SDF or XYZ). Initialize ASE-ANI potential and compute single-point energy for each conformer. Rank conformers by energy and select the lowest-energy subset (retain top N conformers or apply an energy threshold). Export filtered conformer set in a format compatible with downstream quantum-chemical software (QUICK).: "Load conformer structures generated by RDKit in a standard molecular format (e.g., SDF or XYZ). Initialize ASE-ANI potential and compute single-point energy for each conformer. Rank conformers by"
- [other] The workflow uses ASE-ANI for conformation filtering as a step following RDKit conformation generation, operating as part of the automated pipeline for CCS predictions.: "The workflow uses ASE-ANI for conformation filtering as a step following RDKit conformation generation, operating as part of the automated pipeline for CCS predictions."
- [readme] DEPRECATED and no longer supported, please use TorchANI implementation: "DEPRECATED and no longer supported, please use TorchANI implementation"
- [readme] Current ANI-1x and ANI-1ccx potentials provide predictions for the CHNO elements.: "Current ANI-1x and ANI-1ccx potentials provide predictions for the CHNO elements."
- [readme] NOTICE: Python binaries built for python 3.6 and CUDA 9.2. Works only under Ubuntu variants of Linux with a NVIDIA GPU: "NOTICE: Python binaries built for python 3.6 and CUDA 9.2. Works only under Ubuntu variants of Linux with a NVIDIA GPU"
- [readme] workflow allows users to predict CCS values for multiple protonated/deprotonated adducts and models with high automation and parallelized computation on high-performance computing (HPC) systems: "workflow allows users to predict CCS values for multiple protonated/deprotonated adducts and models with high automation and parallelized computation on high-performance computing (HPC) systems"