Efficient Infrared Spectroscopy Using Machine Learning
Author Information
Author(s): Pracht Philipp, Pillai Yuthika, Kapil Venkat, Csányi Gábor, Gönnheimer Nils, Vondrák Martin, Margraf Johannes T., Wales David J.
Primary Institution: Yusuf Hamied Department of Chemistry, University of Cambridge
Hypothesis
This study aims to establish a standard for the efficient computational prediction of IR spectra using a composite approach.
Conclusion
The MACE-OFF23 model provides accurate geometries, frequencies, and IR spectra that outperform traditional methods for drug-like molecules.
Supporting Evidence
- The MACE-OFF23 model shows significantly lower errors compared to GFN methods.
- MACE-OFF23 inherits characteristics similar to DFT methods.
- GFN methods are a reasonable choice for inexpensively calculating frequencies.
- MACE-OFF23 provides excellent performance for molecular geometries and frequencies.
- Classical EEQ models for dipole moment predictions should be avoided.
Takeaway
This study shows how scientists can use computers to quickly and accurately predict how molecules will behave in infrared light, which helps identify unknown compounds.
Methodology
The study combines modern semiempirical quantum mechanical methods with machine learning potentials to predict gas-phase IR spectra.
Limitations
The current implementation of the MACE-OFF23 model does not include molecular electrostatics, limiting its ability to predict dipole moments accurately.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website