Rapid prediction of conformationally-dependent DFT-level descriptors using graph neural networks for carboxylic acids and alkyl amines
2025

Predicting Molecular Descriptors for Carboxylic Acids and Alkyl Amines

Sample size: 16600 publication Evidence: high

Author Information

Author(s): Brittany C. Haas, Melissa A. Hardy, Sowndarya S. V. Shree, Keir Adams, Connor W. Coley, Robert S. Paton, Matthew S. Sigman

Primary Institution: University of Utah

Hypothesis

Can graph neural networks be used to predict DFT-level descriptors for carboxylic acids and alkyl amines?

Conclusion

The study successfully developed models that predict DFT-level descriptors for a large number of carboxylic acids and alkyl amines, significantly reducing computational costs.

Supporting Evidence

  • Models achieved high accuracy (R2 > 0.95) for predicting descriptors.
  • Descriptor libraries were constructed for 8528 carboxylic acids and 8172 alkyl amines.
  • Predictions were validated against an external set of medicinally relevant compounds.
  • 3D models improved accuracy for descriptors sensitive to molecular geometry.

Takeaway

Scientists created a smart computer program that can quickly guess important properties of certain chemicals, making it easier to study them.

Methodology

The study used graph neural networks trained on descriptor libraries to predict molecular descriptors for carboxylic acids and alkyl amines.

Limitations

The models may not capture all stereoelectronic effects due to reliance on 2D representations and the use of inexpensive conformers.

Digital Object Identifier (DOI)

10.1039/d4dd00284a

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication