Inquiry about Adaptations in Retro* for One-Step Model in RetroExplainer

[danielFarfan18]

Hello @wangyu-sd

First, thank you for sharing your work on RetroExplainer. It’s an exciting approach combining chemical knowledge and deep learning for retrosynthesis prediction with quantitative interpretability!

I am particularly interested in the one-step retrosynthesis model used in RetroExplainer. I would like to know if there were any specific adaptations made to the original Retro* code to make it work with your alternate one-step model.

Were there any structural changes to the neural-guided A* search algorithm or other components of Retro* to accommodate RetroExplainer’s one-step approach?