Skip to main content
Fig. 11 | Journal of Cheminformatics

Fig. 11

From: Conditional reduction of the loss value versus reinforcement learning for biassing a de-novo drug design generator

Fig. 11

Property shift demonstration for objective 2. We categorically plotted the distribution of molecules, that is, molecules with optimized properties (number of rings (cycles), number of functional groups, and the R-value), are positioned around x = 1, whereas non-optimized molecules are placed on x = 0. Here we can see another reason why RL-biased-G did not generate many successful molecules for objective 2, which is the inability to design molecules with the required functional groups (second sub-figure on the left), as opposed to the number of the rings and the R-value that were highly optimized only

Back to article page