Fig. 1: Comparison of prior studies and the proposed method. | Communications Chemistry

Fig. 1: Comparison of prior studies and the proposed method.

From: Molecular optimization using a conditional transformer for reaction-aware compound exploration with reinforcement learning

Fig. 1

a Scheme of a rule-based methods. In this example, the ortho-para orientation and steric hindrance are disregarded. b Outline of the previous works. This type of model, which learns a latent space representation, enables smooth compound exploration within the learned distribution. c Our work. A direct tree search occurs from the starting material on the basis of the virtual reactions performed by a conditional transformer model trained on real chemical reactions.

Back to article page