README.md

Data preparation

• configuration files are stored as yaml files under config/prepare_data

Training data configuration file

Single molecule dataset

python3 scripts/prepare_data/prepare_data.py \
  --conf config/prepare_data/train_predictor_mono_config.yaml \
  --data_type mono

Formula MD pretrain dataset

python3 scripts/prepare_data/prepare_data.py \
  --conf config/prepare_data/train_predictor_pretrain_config.yaml \
  --data_type formula

Formula MD+Experimental dataset

python3 scripts/prepare_data/prepare_data.py \
  --conf config/prepare_data/train_predictor_finetune_config.yaml \
  --data_type formula

Formula synthetic dataset

First preprocess the json so the embedding.h5 file can be generated

python3 scripts/prepare_data/prepare_data.py \
  --conf config/prepare_data/synthetic_preprocess_config.yaml \
  --data_type formula

Then run the following command to generate files necessary for diffusion and decoder training

python3 scripts/prepare_data/embedding.py \
  --conf config/prepare_data/synthetic_embedding_config.yaml

You get 3 files from the above command:

• embedding.h5
• solv_embs.pt
• salt_embs.pt

Then preprocess again with the embedding.h5 file for diffusion and decoder training

python3 scripts/prepare_data/prepare_data.py \
  --conf config/prepare_data/train_generator_diffusion_decoder_config.yaml \
  --data_type formula

Inference data configuration file

Single molecule dataset

python3 scripts/prepare_data/prepare_data.py \
  --conf config/prepare_data/inference_predictor_mono_config.yaml \
  --data_type mono

Formula dataset

python3 scripts/prepare_data/prepare_data.py \
  --conf config/prepare_data/inference_predictor_formula_config.yaml \
  --data_type formula