CryoRL

Reinformcement Learning Enables Efficient CryoEM Data Collection

Pre-requisite

Tested in Linux. Please find all package requirements in requirements.txt (tested).

Installation

To create the usable conda environment:
conda create --name <env> --file requirements.txt
Typical installation should take only several minutes.

Dataset

To add new dataset, you should edit the file cryoEM_dataset.py, and use the corresponding files in the folder CryoEM_data (and put new files into it if necessary).

This repo expect you already have the "hole target -- quality metric" pairs ready from the upstream prediction step. For the example format, please see the files in the folder CryoEM_data.

Sample Code

DQN

Train a DQN: python train.py --dataset CryoEM-8bit-resnet18 --lr 0.001 --training-num 10 --test-num 10 --step-per-epoch 500 --seed 2 --ctf-thresh 6 --train-prediction --duration 120 --epoch 10

Note that training would usually take several hours on a typical computer with supported GPU acceleration.

Evaluate: python train.py --dataset CryoEM-8bit-resnet18 --lr 0.001 --training-num 10 --test-num 10 --step-per-epoch 500 --seed 2 --ctf-thresh 6 --train-prediction --duration 120 --epoch 10 --eval

Note that evaluation would usually take less than 1 hour on a typical computer with supported GPU acceleration. This is because we were running many paralell runs on the same dataset with a random starting position.