Fig. 3
From: AI-driven molecular generation of not-patented pharmaceutical compounds using world open patent data
Number of generated molecules matched to the drug-patent DB compounds. Each bar represents the average values of three replicate molecular generations. Error bars represent the standard deviation. In each run, the first 250,000 valid and unique molecules that were generated were evaluated. a Number of patented compounds generated by the patent RNN model and the ChemTS ZINC RNN model under conditions: C = 1.0 and Rrand reward function. These numbers are used as baseline values of patented compound-generations for the RNN models. b Number of patented compounds generated by the patent RNN model across varying reward functions. The baseline values of the patent/ZINC RNN models are also shown as dashed lines