① Input Type

Single SMILE SMILES File

No SMILES in mind? Try one of these: Dopamine, Oxytocin, Glucose, Serotonin, Melatonin Have a formula but no SMILE? Click here: SMILX-isogenerator

② 3D Construction Method 🛈

③ Force Field optimization (Optional) 🛈

④ Post-optimization (Advanced) 🛈

3D Visualization will appear here

Welcome to ELAYA SMILES

Energy-minimized Linear-to-structure Atom Yielding Algorithm

A web platform that converts SMILES notation into optimized 3D molecular geometries — interactively and for free.

View on GitHub

The notation

What is a SMILES string?

SMILES — Simplified Molecular Input Line Entry System — is a compact text-based notation for encoding molecular structure. Each character encodes atoms, bonds, branches, rings, and stereochemistry.

It is the standard exchange format in databases such as PubChem, ChEMBL, and ZINC — the starting point for any QM/MM or docking pipeline.

Molecule	SMILES	Notes
Benzene	`c1ccccc1`	Aromatic ring
Dopamine	`C1=CC(=C(C=C1CCN)O)O`	Catecholamine
Serotonin	`C1=CC2=C(C=C1O)NC=C2CCN`	Indole scaffold
Glucose	`OC[C@H]1O[C@@H](O)…`	Stereocenters (@)
Oxytocin	`CC(C)C[C@H](NC(=O)…`	Nonapeptide

Encoding Rules

C / c

Aliphatic / aromatic carbon

= #

Double / triple bond

( )

Branch point

1…9

Ring closure number

@ @@

Tetrahedral stereo (R/S)

[NH3+]

Charged atom in brackets

/ \

Double-bond geometry (E/Z)

The science

Why does 3D geometry matter?

A SMILES string describes how atoms are connected — but chemistry happens in three dimensions. Biological activity, receptor binding, spectroscopic signatures, and reaction pathways are all governed by the precise spatial arrangement of atoms.

Generating an accurate 3D structure is often a critical first step before molecular dynamics simulations, docking studies, quantum-chemical calculations, or certain machine-learning models that rely on structural molecular features.

Conversion methods

Three algorithms, one interface

ELAYA offers three complementary 3D construction methods, each suited to different use cases.

RDKit

Distance-geometry embedding (ETKDGv3) followed by force-field minimization.

OpenBabel

Template-based coordinate builder. Offers broader chemical coverage than many toolkits, though complex systems such as organometallics may still require specialized methods for accurate geometries.

NetworkX

Graph-theoretic spring layout. Fast and lightweight, useful for topological exploration.

RDKit

Accuracy

OpenBabel

Coverage

NetworkX

Speed

Deep dive

RDKit — ETKDGv3 pipeline

The most precise method implemented in ELAYA

SMILES parsing Chem.MolFromSmiles(smiles)

Builds the molecular graph: nodes = atoms, edges = bonds. Validates SMILES syntax.

Explicit hydrogens Chem.AddHs(mol)

Adds all hydrogens explicitly — required for correct 3D geometry.

3D seed (ETKDGv3) AllChem.EmbedMolecule(mol, ETKDGv3())

Experimental Torsion Knowledge Distance Geometry v3 (ETKDGv3). Incorporates knowledge-based torsional preferences derived from experimental and computational data to generate realistic 3D conformations. A fixed random seed (665) is used for reproducibility.

FF optimization UFFOptimizeMolecule / MMFFMinimize

Minimizes energy with UFF or MMFF94 (max 2000 iterations). If MMFF94 fails, falls back to UFF automatically.

Export → XYZ + MOL block

Generates the XYZ file (N atoms, SMILES comment, coordinates) and a MOL block for the 3Dmol.js viewer.

Energy minimization

Force field optimization

A set of mathematical equations and parameters that model forces between atoms: bond stretching, angle bending, dihedral torsion, and non-covalent interactions (van der Waals, electrostatics). Minimizing energy with these equations produces more realistic 3D geometries than a bare embedding.

UFF

Universal Force Field

RDKitOpenBabel

General — covers the entire periodic table. Based on atomic hybrid geometries.

MMFF94

Merck Molecular Force Field

RDKit

Parametrized for drug-like organic molecules. High precision in bond geometry and angles.

Fallback behavior: in OpenBabel, if the force field name is not found in the system, optimization is skipped and raw OBBuilder coordinates are used. In RDKit, if MMFF94 fails for a molecule, it automatically falls back to UFF.

Advanced search

Conformational search with GLOMOS

GLOMOS (Genetic-algorithm-based Local Optimization of MOlecular Structures) implements a heuristic genetic algorithm to explore the conformational space of rotatable bonds using ANI-based neural network potentials. It is integrated as a streaming endpoint (/api/convert/stream) with real-time Server-Sent Events (SSE) feedback.

Most molecules exist as a family of interconverting 3D shapes — conformers — differing in the rotation of single bonds. GLOMOS aims to identify low-energy conformers with accuracy approaching quantum-chemical methods for many organic systems, while avoiding the high computational cost of DFT calculations.

Parameter	Default	Description
`initpop`	4	Initial population size
`matings`	2	Crossovers per generation
`mutants`	2	Random mutations per generation
`generations`	3	Total number of generations
`energy_cutoff`	4.0	Energy window (kcal/mol)
`tol_similarity`	0.95	Similarity tolerance between conformers
`cutoff_population`	8	Max population size
`nof_repeats`	2	Repeats before stopping
`nof_stagnant`	3	Stagnant generations → stop
`ani_model`	ANI1ccx	Neural potential (ANI1x / ANI1ccx / ANI2x)

Implementation

Genetic algorithm pipeline & real-time streaming

Initialize

Seed 3D geometry from SMILES via RDKit

→

Population

Random rotamer variants sampled from torsion angles

→

Selection

ANI potential evaluates energy of each conformer

→

Crossover

Mating and mutation of best rotamer candidates

→

Convergence

Best conformer delivered after N generations

Real-time SSE events emitted to the browser

startEmits total number of generations

logEach stdout line from the subprocess

progressCurrent generation, min energy, conformer name

doneXYZ of the best conformer found

errorError message if the process fails

Workflow

From string to structure in five steps

Input

Enter a SMILES string or upload a batch file with one SMILES per line.

→

Configure

Select an algorithm (RDKit, OpenBabel, or NetworkX) and an optional force field.

→

Optimize

Optionally run GLOMOS to sample low-energy conformers with ANI potentials.

→

Visualize

Inspect the interactive 3D structure or switch to a 2D diagram at any time.

→

Export

Download in XYZ, SDF, MOL, or MOL2 format for use in external software.

See it in action

Live conversion of a SMILES string into a fully optimized 3D structure

Technologies

Backend	Flask · Python 3
Cheminformatics	RDKit · OpenBabel · DScribe
Neural potentials	ANI-1ccx / ANI-2x via TorchANI
Visualization	3Dmol.js · Chart.js
Frontend	HTML5 · CSS3 · Vanilla JavaScript

Project Authors

ELAYA SMILES was designed and developed by Gabriela Vidales, with scientific guidance from MSc Luis Ortiz, PhD Filiberto Ortiz Chi and PhD Gabriel Merino.

Passion for building tools that bring science closer to people. Fostering open and reproducible science.

View on GitHub ↗

ELAYA SMILES

"Convert SMILES to 3D/2D structure"

Welcome to ELAYA SMILES

What is a SMILES string?

Why does 3D geometry matter?

Three algorithms, one interface

RDKit — ETKDGv3 pipeline

Force field optimization

Conformational search with GLOMOS

Genetic algorithm pipeline & real-time streaming

From string to structure in five steps

Technologies

Project Authors

Share your feedback

Bug report

Suggestion

Contribute