The following repository consists of a concise version of the backend code for the publication:
Classification of substances by health hazard using deep neural networks and molecular electron densities. by Singh et al., 2024
The data required for the ECHA dataset is present in the following files:
- Training_data_ECHA_unsorted.csv
- Test_data_ECHA_unsorted.csv
- Val_data_ECHA_unsorted.csv
The directory format required for the cube files is shown in the directory ECHA_cubes and tox_datasets.
The dataloaders for the two datasets are written in open_foodtox_dataloader.py and custom_loader.py.
As shown in the code, the files are expected to have a certain nomenclature. For eDen files, this is expected to be: index_e_Dens_vol_data_.npy and for ENeg files, it's expected to be index_e_Eneg_vol_data_.npy where index refers to the integer index of the files, starting at 0.
Moreover, all the voxel shapes for the files are expected in a mode_vox_info.npy file where mode refers to either: train, test or val.
The training (and optimisation) loops are present in open_foodtox_optimise.py and Optuna_tuned_Echa.py.
The SMILES strings for both datasets are present in files: CompFood.csv and ECHA csv files split into each of the train/test/val sets with allowed or prohibited labels.