Skip to content

Latest commit

 

History

History
129 lines (81 loc) · 3.69 KB

README.md

File metadata and controls

129 lines (81 loc) · 3.69 KB

SD VAE

Background

Deep generative models are rapidly becoming popular tools for generating new molecules and optimizing the chemical properties. In this work, we will introduce a VAE model based on the grammar and semantic of molecular sequence - SD VAE.

Instructions

This repository will give you the instruction of training a SD VAE model.

Data Link

You can download the data from [datalink] (https://baidu-nlp.bj.bcebos.com/PaddleHelix/datasets/molecular_generation/data_SD_VAE.tgz).

  1. create a folder './data'
  2. unzip the file into './data'

data (project root)

|__ data_SD_VAE

|__ |__ context_free_grammars

|__ |__ zinc

Data pre-processing

Before training/evaluation, we need to cook the raw txt dataset.

cd data_preprocessing

python make_dataset_parallel.py \
-info_fold ../data/data_SD_VAE/context_free_grammars \
    -grammar_file ../data/data_SD_VAE/context_free_grammars/mol_zinc.grammar \
    -smiles_file ../data/data_SD_VAE/zinc/250k_rndm_zinc_drugs_clean.smi 
    
    
python dump_cfg_trees.py \
-info_fold ../data/data_SD_VAE/context_free_grammars \
    -grammar_file ../data/data_SD_VAE/context_free_grammars/mol_zinc.grammar \
    -smiles_file ../data/data_SD_VAE/zinc/250k_rndm_zinc_drugs_clean.smi

The above two scripts will compile the txt data into binary file and cfg dump, correspondingly.

Training

Model config

The model config is the parameters used for building the model graph. They are saved in the file: model_config.json

"latent_dim":the hidden size of latent space
"max_decode_steps": maximum steps for making decoding decisions
"eps_std": the standard deviation used in reparameterization tric
"encoder_type": the type of encoder
"rnn_type": The RNN type

Training setting

In order to train the model, we need to set the training parameters. The dafault paramaters are saved in file: args.py

-loss_type : the type of loss
-num_epochs : number of epochs
-batch_size : minibatch size
-learning_rate : learning_rate
-kl_coeff : coefficient for kl divergence used in vae
-clip_grad : clip gradients to this value

To run the trianing scripts:

CUDA_VISIBLE_DEVICES=0 python train_zinc.py \
-mode='gpu' \

Download the trained-model

You can download the trained model from (https://baidu-nlp.bj.bcebos.com/PaddleHelix/models/molecular_generation/SD_VAE_model.tgz).

unzip the file and put the model into './model' folder:

|__ model

|__ |__ train_model_epoch499

Run sampling

Sample from normal distribution prior:

python sample_prior.py \
  -info_fold ../data/data_SD_VAE/context_free_grammars  \
  -grammar_file ../data/data_SD_VAE/context_free_grammars/mol_zinc.grammar \
  -model_config ../model_config.json \
  -saved_model ../model/train_model_epoch499

reconstruct from the reference sequence:

python reconstruct_zinc.py  \
  -info_fold ../data/data_SD_VAE/context_free_grammars \        
  -grammar_file ../data/data_SD_VAE/context_free_grammars/mol_zinc.grammar \      
  -model_config ../model_config.json \       
  -saved_model ../model/train_model_epoch499 \
  -smiles_file ../data/data_SD_VAE/zinc/250k_rndm_zinc_drugs_clean.smi
Sampling results from prior

valid: 0.49

unique@100: 1.0

unique@1000: 1.0

IntDiv: 0.92

IntDiv2: 0.82

Filters: 0.30

Reconstruction result

accuracy: 0.92

Reference

[1] @misc{dai2018syntaxdirected, title={Syntax-Directed Variational Autoencoder for Structured Data}, author={Hanjun Dai and Yingtao Tian and Bo Dai and Steven Skiena and Le Song}, year={2018}, eprint={1802.08786}, archivePrefix={arXiv}, primaryClass={cs.LG} }