FIGURE 2.

(i) Some GAI models; (ii) De novo molecular design workflow (involving transfer learning to address data scarcity) for generating small molecules against a pre-decided target. (A) Pre-trained deep generative model. (B) TL to learn the essential characteristics of small compounds concerning the target receptor, and RL approach to enhance the molecular properties in the desired direction. (C) Various physicochemical property filters, structural alerts, and virtual screening scores for selecting or ranking model-generated molecules for the next level.