How to use property constraints for Mol2Mol sampling? #142
-
|
I am trying to understand how to use Mol2Mol that generate molecules that satisfy given property constraints. By reading the Reinvent 4 paper, this should be possible. Table 4 of the paper lists Mol2Mol models originally described in paper Transformer-based molecular optimization beyond matched molecular pairs. These Mol2Mol variants should accept as input a property constraint (alongside the SMILES sequence), and generate new SMILES that satisfy the property constraint. I read all the examples provided in this repository, and I searched the source code, but I don't understand how I can specify the property constraint in the configuration for a sampling run. Please let me know if it's possible to use Mol2Mol in sampling mode while specifying property constraints. If I misunderstood the papers and this feature is not supposed to exist, please also let me know, I really want to understand how Mol2Mol works. Thanks, |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
Hi, many thanks for your interest in REINVENT and welcome to the community! The prior you are referring to is not available in REINVENT and I suggest to directly contact the authors of that paper. REINVENT uses reinforcement learning (RL) to restrain molecule generation to desired property. Multiple scoring components are readily available in the current release. Many thanks, |
Beta Was this translation helpful? Give feedback.
Hi,
many thanks for your interest in REINVENT and welcome to the community!
The prior you are referring to is not available in REINVENT and I suggest to directly contact the authors of that paper. REINVENT uses reinforcement learning (RL) to restrain molecule generation to desired property. Multiple scoring components are readily available in the current release.
Many thanks,
Hannes.