Skip to main content
Fig. 5 | Journal of Cheminformatics

Fig. 5

From: Augmented Hill-Climb increases reinforcement learning efficiency for language-based de novo molecule generation

Fig. 5

Optimization of normalized docking score and uniqueness during optimization across three targets. (a) With diversity filter 1 (DF1), docking score converges to the minimum score threshold (0.8) of the DF and model undergoes mode collapse seen by an associated drop in uniqueness. (b) With diversity filter 2 (DF2), no convergence is observed, and uniqueness maintains relatively high. This is due to a lower minimum score threshold (0.5) and softer penalization scheme

Back to article page