zjuKeLiu/Mol-GenDA

StructureGeneration

JT-VAE

• JT-VAE is the decoder in this work
• From: https://github.com/wengong-jin/icml18-jtnn/tree/370c6d1a9fe6ed44841430de2febee36f694678d

GAN

• Generator

one-shot GenDA

• Domain adaption

Usage

• In ./GAN

  • "sh train.sh" train the GAN with the whole dataset ZINC-250K

  • "sh finetune.sh" finetune the GAN with specific dataset so that it can generate materials similar to the target

  • "sh generate.sh" generate the desired material

• In ./GEN

  • "sh decode.sh" docode the material SMILEs from latent vector

• In ./Valid

  • "sh valid.sh" validate the results