Skip to main content
. 2023 Oct 3;14:6155. doi: 10.1038/s41467-023-41698-5

Fig. 6. Distributions of t-SNE from hidden layers of RetroExplainer.

Fig. 6

ad Hidden features are derived based on two criteria: (1) whether the reaction type is recognized, and (2) the origin of the hidden features (either from the reaction center predictor, RCP, layer, or the leaving group matcher, LGM, layer). Subsequently, these features are reduced to two dimensions using t-SNE techniques. Distinct colors represent various reaction types, while diverse styles are assigned based on the maximum count of hydrogen alterations.