Auto3D

Auto3D generates low-energy three-dimensional conformers from SMILES strings to provide accurate molecular geometries for computational chemistry and chemical discovery.

Key Features:

• Automated Stereoisomer Enumeration: Identifies and enumerates possible stereoisomers to cover potential stereochemical configurations.
• Duplicate Filtering: Filters out duplicate structures to retain unique molecular candidates.
• 3D Structure Building: Constructs three-dimensional models directly from SMILES inputs.
• Fast Geometry Optimization: Performs rapid geometry optimizations to refine structures toward low-energy states.
• Ranking Process: Ranks generated conformers by stability and energy using neural network atomistic potentials.

Scientific Applications:

• Stereoconfiguration Identification: Evaluated on 50 molecules with multiple stereocenters to identify the stereoconfiguration yielding the lowest-energy conformer.
• Tautomeric Reaction Energy Benchmarking: Integration with ANI-2xt yields a 42% reduction in error for tautomeric reaction energy calculations compared with ANI-2x when benchmarked against coupled-cluster reference data.

Methodology:

Auto3D applies neural network atomistic potentials (ANI and ANI-2xt), with ANI-2xt trained on a tautomer-rich dataset, to predict energies, perform geometry optimizations, and rank conformers, achieving optimizations faster than traditional Density Functional Theory (DFT) methods.