a simple vae to encode protein condensate + small molecule trajectory data

requirements

• numpy
• pandas
• tqdm
• sklearn
• pytorch
• torchvision
• pyro
• seaborn

data format

Data format is expected to be a csv with each row as a datapoint. If your data has a label column and you'd like to exclude it from the representation, pass in label_col=NAME_OF_COL to the InteractionDataset initialization in trainer.py. The default label_col is protein.

training

In train.py,

1. update DATA_DIR and MODEL_DIR to be compatible with your directory setup.
2. run python train.py --input_dim 239 --z_dim 16 --hidden_dim 64 or whatever your desired dimensions are.

If you'd like to use a slurm submission script (particularly with MIT's SuperCloud), edit the llsub.sh file and try running LLsub llsub.sh -g volta:1. If you are using a conda environment, change the name (mine is called pcsm).

evaluation

An example of running evaluation is included in llsub.sh.