Data and Code for :Constrained SELFIES-Based Inverse Design of Ionic Liquids for Fuel Denitrogenation
Description
This dataset contains the input workbooks and Jupyter notebooks used to reproduce the constrained and unconstrained SELFIES-based ionic-liquid generation and ranking workflows. density_P.xlsx and viscosity_P.xlsx provide the literature-derived ion records and family labels used to construct the parent/seed ion pool. Their measured density and viscosity values were not used as generation or ranking targets. All_contaminants_decane_2_with_ion_smiles_CHARGE_OK.xlsx contains charge-checked ion SMILES and precomputed COSMOtherm data for the decane, quinoline, and carbazole ranking tasks. Constrained_mutations_and_relevant_codes.ipynb implements constrained SELFIES mutation, structural filtering, PubChem screening, two-stage learning-to-rank, deployment, and the associated robustness and generalization analyses. Unconstrained_mutation_ablation_and_related_codes.ipynb implements the relaxed comparison workflow together with the sampling and structural-complexity analyses. The notebooks were run in Google Colab, with an NVIDIA A100 GPU used for model training. File paths may require updating. Independent COSMO-RS recalculation requires licensed BIOVIA COSMOtherm software.
Files
Institutions
- Khalifa University of Science and TechnologyAbu Dhabi, Abu Dhabi