Official GitHub repository for LDMol, a latent text-to-molecule diffusion model. The details can be found in the paper LDMol: Text-Conditioned Molecule Diffusion Model Leveraging Chemically Informative Latent Space.

LDMol not only can generate molecules according to the given text prompt, but it's also able to perform various downstream tasks including molecule-to-text retrieval and text-guided molecule editing.

The unavoidable discreteness of a molecule makes it difficult for a diffusion model to connect raw data with highly complex conditions like natural language. To address this, we present a novel latent diffusion model dubbed LDMol for text-conditioned molecule generation. LDMol comprises a molecule autoencoder that produces a learnable and structurally informative feature space, and a natural language-conditioned latent diffusion model. In particular, recognizing that multiple SMILES notations can represent the same molecule, we employ a contrastive learning strategy to extract feature space that is aware of the unique characteristics of the molecule structure. LDMol outperforms the existing baselines on the text-to-molecule generation benchmark, suggesting a potential for diffusion models can outperform autoregressive models in text data generation with a better choice of the latent domain. Furthermore, we show that LDMol can be applied to downstream tasks such as molecule-to-text retrieval and text-guided molecule editing, demonstrating its versatility as a diffusion model.

Run conda env create -f requirements.yaml and it will generate a conda environment named ldmol.

The model checkpoint and data are too heavy to be included in this repo and can be found in here.

Check out the arguments in the script files to see more details.

The model performs molecule-to-text retrieval on the given dataset. --level controls the quality of the query text(paragraph/sentence). --n-iter is the number of function evaluations of our model.

The model performs a DDS-style text-guided molecule editing. --source-text should describe the --input-smiles. --target-text is your desired molecule description.