Fig. 5

Docking scores (a) and predicted probability of DRD2 activity (b) of molecules generated de novo using the Prior, the SVM-Agent and the Glide-Agent, compared to the active, inactive, and random reference datasets. The more negative the docking score, the better it is predicted to bind. The Glide-Agent generated molecules have the best docking score distribution, more so than known DRD2 active molecules, whilst the SVM-Agent generated molecule distribution is more similar to known DRD2 active molecules. The SVM-Agent molecules and known DRD2 actives score most highly according to the SVM, comparatively, the Glide-Agent molecules do not