Improving the reliability of molecular string representations for generative chemistry

This zenodo repository contains the dataset used for training models described in : "Improving the reliability of molecular string representations for generative chemistry",2025, Reboul et al. It also includes the checkpoints of VAE models weights each five epochs and samples generated from the final models. Training_data : Cointains the processed molecular sets (MOSES) datasets ( train and test sets) : SMILES: canonical SMILES using RDKIT 2023.09.1, sampled Clearsmiles ( cf. paper) SELFIES: SELFIES using python module selfies 2.1, SELFIES with no overload and SELFIES with no hexadecimal encoding Checkpoints : xxx_yyy_modelname_lat_zz.ckpt where xx is the current epoch saved , yyy the maximum epoch during training, zz the number of latent dimensions in latent space Samples : csv or parquet files , samples_xxx_yyy_modelname_late_zz , xxx,yyy, and zz are the same as above.