macrocycles

[indazole]

Hello,
I have been trying out GDSS, I think it works well - great job!
One observation, I notice the sampled 10K compounds usually contain many macrocycles and large ringed molecules even though the training set has few. Depending on the desired use, this could be good or not so good.
Do you have an idea on how to minimize the numbers of macrocycles sampled, eg: more training epochs, etc ?
Cheers,

[harryjo97]

Hi,
Thanks for your interest in our work!
We also observed that some of the generated molecules had unusually large rings.

In order to generate molecules that satisfy constraints, we present a guided molecule generation framework based on GDSS in
Lee et al,. Exploring Chemical Space with Score-based Out-of-distribution Generation, ICML 2023 (https://arxiv.org/abs/2206.07632)
You can apply a cycle condition to generate more stable molecules!

[indazole]

Thanks for the reply. I will look at MOOD asap!