Language Selection

• English
• 中文

Pre-print Paper

The pre-print paper describing the model can be found on ChemRxiv: Prediction of Molecular Critical Properties Based on Boruta Feature Selection: Balancing Accuracy, Applicability and Explainability.

BiBTeX entry for citation:

@article{Li2026CriticalProperties,
      author = {Xurui Li  and Zhiguo Gan  and Jiaming Zhang  and Hongxi Zeng  and Zheng Liu  and Diannan Lu },
      title = {Prediction of Molecular Critical Properties Based on Boruta Feature Selection: Balancing Accuracy, Applicability and Explainability},
      journal = {ChemRxiv},
      volume = {2026},
      number = {0401},
      pages = {},
      year = {2026},
      doi = {10.26434/chemrxiv.15001485/v1},
      URL = {https://chemrxiv.org/doi/abs/10.26434/chemrxiv.15001485/v1},
      eprint = {https://chemrxiv.org/doi/pdf/10.26434/chemrxiv.15001485/v1}
}

Repository Introduction

• This repository is for the paper Prediction of Molecular Critical Properties Based on Boruta Feature Selection: Balancing Accuracy, Applicability and Explainability. It contains related contents such as test set data, models, and test results.
• Because the ThermoData Engine (TDE) is a commercial database, this study is for academic purposes only. Without the permission of TDE, the critical data points of the test dataset cannot be published. Therefore, the test dataset only contains molecular names and SMILES strings, and the experimental critical data points have been manually hidden. Since we provide the original calculation code, the AARD may appear as 'nan' during the calculation due to the missing data, but this does not affect the normal execution of the code.

Folder Description

• The data folder contains feature normalization parameters.
• The homo folder contains the prediction results of critical points for homologues.
  • PURE-homo-150.xlsx contains four classes of PONA hydrocarbon homologues, with 150 points for each class.
• The image folder contains the performance plots of the models on the training and validation sets.
• The model folder contains the parameters of the trained models.
• The newexp folder contains the prediction performance of the models tested on new substances.
  • TEST_A.xlsx contains the test data for Test(A).
  • TEST_AB.xlsx contains the test data for Test(A) and Test(B).

Description of .ipynb Files

• GNN_predict.ipynb uses the GNN model to make predictions on Test(A).
• GRP_Pc_2Dpredict.ipynb uses the GRP (2D) model to predict $P_c$ on Test(A). The internal outputs are the prediction metrics on Test(A), Test(ALL), and Test(B), respectively.
• GRP_Tc_2Dpredict.ipynb uses the GRP (2D) model to predict $T_c$ on Test(A). The internal outputs are the prediction metrics on Test(A), Test(ALL), and Test(B), respectively.
• RF_Pc_2Dpredict.ipynb uses the RF model to predict $P_c$ on Test(A). The internal outputs are the prediction metrics on Test(A), Test(ALL), and Test(B), respectively.
• RF_Tc_2Dpredict.ipynb uses the RF model to predict $T_c$ on Test(A). The internal outputs are the prediction metrics on Test(A), Test(ALL), and Test(B), respectively.
• SVR_Pc_2Dpredict.ipynb uses the SVR model to predict $P_c$ on Test(A). The internal outputs are the prediction metrics on Test(A), Test(ALL), and Test(B), respectively.
• SVR_Tc_2Dpredict.ipynb uses the SVR model to predict $T_c$ on Test(A). The internal outputs are the prediction metrics on Test(A), Test(ALL), and Test(B), respectively.
• predict_3D.ipynb contains the prediction code for all 3D models on the test set. It internally outputs the prediction metrics of the GPR, RF, and SVR models on Test(ALL) and Test(A), respectively.

Test Data Description

• All test sets are derived from the ThermoData Engine (TDE) database and consist of two parts: Test(A) and Test(B).
• Test(A) contains 148 molecules, and Test(B) contains 38 molecules.
• The entire test set was obtained through manual individual queries, as the TDE software does not allow batch queries or batch data exports.
• All data in the test set use the critical data points evaluated and accepted by TDE. If there are multiple experimental data points, their average value is calculated and used.
• Because TDE is a commercial database, this study is for academic purposes only. Without the permission of TDE, the critical data point values of the test dataset cannot be published.