Skip to main content


Fig. 5 | Journal of Cheminformatics

Fig. 5

From: An exploration strategy improves the diversity of de novo ligands using deep reinforcement learning: a case for the adenosine A2A receptor

Fig. 5

The value of the loss function and the percentage of valid SMILES sequences during the pre-training process on the ZINC set (a) and fine-tuning process on the A2AR set (b). The model was well pre-trained after 300 epochs and these two values converged to 0.19 and 93.88%, respectively. The performance of the fine-tuned model converged after 400 epochs with the two values reaching 0.09 and 99.73%, respectively

Back to article page