Fig. 3
From: LOGICS: Learning optimal generative distribution for designing de novo chemical structures
Chemical space visualization by t-SNE. Chemical space visualization of the Morgan fingerprint vectors transformed by t-SNE on (a) the KOR experiment and (b) the PIK3CA experiment, respectively. The t-SNE components were calculated with the test set actives (red) and the valid molecules among generations from the prior model as well as the different GPC models (blue). For the purpose of visualization, 2,000 randomly selected, valid, generated molecules were used for each generative model