Fig. 4
From: InertDB as a generative AI-expanded resource of biologically inactive small molecules from PubChem
Deep generative AI model for proposing potential inactive compounds. a, Generative AI for producing potential inactive compounds from CICs. b, A schematic diagram describing input and output of RNN-based generative model. c. SMILES augmentation. d,e. Performances of generative AI for fraction valid (d) and fraction novel (e) among generated SMILES varying the augmentation factor and the number of RNN layers. f, Chemical space of CICs and pGICs. g, Nearest neighbor chemical similarity (Tc) to CICs by pGICs with different generating frequencies. h, Proportion of pGICs found in PubChem across different generating frequency subsets. i, Cumulative distribution of the fraction of active assay results for individual compounds in different pGIC subsets