<ul data-eligibleForWebStory="true"><li>Generative Flow Networks (GFlowNets) are used for molecular graph generation.</li><li>Previous methods restricted exploration by using predefined molecular fragments.</li><li>Atomic GFlowNets (A-GFNs) introduce a new generative model using individual atoms as building blocks for drug-like chemical space exploration.</li><li>Unsupervised pre-training with drug-like molecule datasets is proposed for A-GFNs, focusing on molecular descriptors like drug-likeliness and synthetic accessibility scores as rewards.</li><li>These rewards guide A-GFNs towards regions of chemical space with desired pharmacological properties.</li><li>Goal-conditioned finetuning helps adapt A-GFNs for specific target properties.</li><li>Pretraining A-GFN on a subset of ZINC dataset shows effectiveness in drug design tasks compared to baseline methods.</li><li>The code for A-GFN is available at https://github.com/diamondspark/AGFN.</li></ul>

Pretraining Generative Flow Networks with Inexpensive Rewards for Molecular Graph Generation

Discover more