De novo, in-silico design of molecules is a challenging problem with applications in drug discovery and material design. In our work, we introduce a masked graph model, which learns a distribution over graphs by capturing conditional distributions over unobserved nodes (atoms) and edges (bonds) given large sets of observed molecules. We train our model on existing molecular graphs and then sample novel ones from it by iteratively masking and replacing different parts of initialized graphs.<br/><br/>We evaluate our approach on the QM9 and ChEMBL datasets using the GuacaMol framework's distribution-learning benchmark. The benchmark contains five metrics: the validity, uniqueness, novelty, KL-divergence and Fr{\'e}chet ChemNet Distance scores, the last two of which are measures of the similarity of the generated samples to the training, validation and test distributions. We find that validity, KL-divergence and Fr{\'e}chet ChemNet Distance scores are anti-correlated with novelty, and that we can trade off between these metrics more effectively than existing models. We observe that our model outperforms previously proposed graph-based approaches and is competitive with SMILES-based approaches.<br/><br/>Finally, we show that our model can generate molecules with desired values of specified properties while maintaining physiochemical similarity to the training distribution.

Masked graph modeling for molecule generation

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.