Hidden_vs_Desc

Code and data used in the paper: Harnessing Surrogate Models for Data-efficient Predictive Chemistry: Descriptors vs. Learned Hidden Representations

Files Organization

• Energy_HAT: our first case study Activation Energy Prediction for HAT Reactions.

• Selectivity: our second case study Regio-selectivity Prediction.

• Non_Reactivity: our third case study for a range of Molecular Property Predictions.

• ROGI_XD: our toy model for the roughness index ROGI-XD.

Notes

• Certain checkpoints of trained models are managed through Git Large File Storage.

• For more details and code environments, please see the individual readme in each folder. Or see the respective repositories of Alfonso-Ramos et al., Guan et al., and Li et al.

• In each case study, a function mean_std() is provided in the data_manager.py to help you organize the results and get mean values and standard deviations from different random seeds.

• The code is tested on a server with:

  • Linux Operating System
  • NVIDIA A30 GPU
  • 64-core CPU