Please first unzip the compressed files for all 12 models. For each model (e.g., density_GAFF_M1), the descriptor set and model category are indicated in the directory name:

density_GAFF_M1
density_GAFF_M2
density_GAFF_M3
density_RDKIT_M1
density_RDKIT_M2
density_RDKIT_M3
nD_GAFF_M1
nD_GAFF_M2
nD_GAFF_M3
nD_RDKIT_M1
nD_RDKIT_M2
nD_RDKIT_M3

We constructed three categories of models using two types of molecular descriptors: GAFF and RDKIT.

M1 models were trained using all available data.
M2 models included temperature as an input feature during training.
M3 models were trained using datasets containing temperature information, although temperature itself was not used as a training feature.

Each model directory contains:
Training and test datasets in CSV format
Three MATLAB code files:

retrain.m — retrains the model over 20 independent runs
prediction.m — performs predictions using the best-performing model selected from the 20 runs and saves the resulting MATLAB data file
trainRegressionModel.m — function called by retrain.m for model training

train_log.docx: the performance of 20 independent runs, with the summary of the best model highlighted in yellow. 

The trained model is stored as a MATLAB data file (.mat) and can be loaded directly into MATLAB. After loading, the trained model object stored in the variable model_name can be used for prediction. Associated model parameters can also be inspected within the MATLAB workspace. The variable validationRMSE provides the cross-validation RMSE of the model.
To use a trained model, navigate to the corresponding model directory, load the MATLAB data file into MATLAB, and execute prediction.m to generate predictions. The code may also be modified to perform predictions on new datasets.
Due to the extremely large file size, the best-performing model file (bestmodel.mat) must be downloaded separately via

web browser: 
https://mulan.pharmacy.pitt.edu/group/github/density-nD/density_GAFF_M1/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/density_GAFF_M2/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/density_GAFF_M3/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/density_RDKIT_M1/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/density_RDKIT_M2/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/density_RDKIT_M3/bestmodel.mat  

https://mulan.pharmacy.pitt.edu/group/github/density-nD/nD_GAFF_M1/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/nD_GAFF_M2/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/nD_GAFF_M3/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/nD_RDKIT_M1/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/nD_RDKIT_M2/bestmodel.mat  
https://mulan.pharmacy.pitt.edu/group/github/density-nD/nD_RDKIT_M3/bestmodel.mat  

After download, bestmodel.mat must be placed in the corresponding directory, e.g.
bestmodel.mat downloaded from https://mulan.pharmacy.pitt.edu/group/github/density-nD/density_GAFF_M1/bestmodel.mat  
must be moved to density_GAFF_M1

One can also obtain the models using wget by running download_model.sh