This repository basically reproduce the result of Learning Multimodal Graph-to-Graph Translation for Molecular Optimization (ICLR 2019)
Original Repository: https://github.com/wengong-jin/iclr19-graph2graph
The current method utilizes basic dot-product attention, and we tried the scaled dot-product because the scaling factor would promote more efficient learning since dot-product grows large when input is large, thus leading vanishing gradients in the softmax function which normalizes the attention score.
A quick summary of different folders:
data/contains the training, validation and test set of logP, QED and DRD2 tasks described in the paper.data_processing_pyspark/contains the implementation of pyspark to process raw data (README).diff_vae/includes the training and decoding script of variational junction tree encoder-decoder (README).diff_vae_gan/includes the training and decoding script of adversarial training module (README).fast_jtnn/contains the implementation of junction tree encoder-decoder.props/is the property evaluation module, including penalized logP, QED and DRD2 property calculation.scripts/provides evaluation and data preprocessing scripts.
Muyang Sun, Chong Dang, Fangxiang Wang, Jingliang Zhang