Auto3D v2.2 Released with AIMNet2 Integration
Major update brings faster conformer generation powered by AIMNet2 neural network potentials
Announcement
Auto3D v2.2 Released with AIMNet2 Integration
We are excited to announce the release of Auto3D v2.2, featuring full integration with AIMNet2 neural network potentials for faster and more accurate 3D molecular structure generation.
What’s New
AIMNet2 as Default Engine
Auto3D now uses AIMNet2 as the default neural network potential, replacing the previous ANI models. This brings significant improvements:
- Broader Coverage: Support for 14 elements covering >90% of drug-like molecules
- Charged Species: Accurate handling of ions and charged molecules
- Improved Accuracy: Better energy rankings for conformer selection
Performance Improvements
- 5× faster geometry optimization with GPU acceleration
- Lower memory footprint for batch processing
- Parallel processing for high-throughput workflows
New Features
- Tautomer enumeration and ranking
- Improved command-line interface
- Better integration with RDKit and ASE
- Support for periodic boundary conditions
Installation
pip install --upgrade auto3dQuick Example
from auto3d import Auto3D
# Generate conformers from SMILES
result = Auto3D("CCO") # Ethanol
conformers = result.get_conformers()
print(f"Found {len(conformers)} low-energy conformers")Resources
Acknowledgments
This work was supported by NSF and NIH grants. We thank all contributors and users who provided feedback to improve Auto3D.
Try the new Auto3D and let us know what you think!