CryoPROS is a computational framework specifically designed to tackle misalignment errors caused by preferred orientation issues in single-particle cryo-EM. It addresses these challenges by co-refining synthesized and experimental data. By utilizing a self-supervised deep generative model, cryoPROS synthesizes auxiliary particles that effectively eliminate these misalignment errors through a co-refinement process.

References:

• Zhang, H., Zheng, D., Wu, Q. et al. CryoPROS: Correcting misalignment caused by preferred orientation using AI-generated auxiliary particles Nat Commun 16, 4565 (2025).

• CryoPROS Github page wtih detailed workflow: (Copy the tutorial data from ${CRYOPROS_EXAMPLES} on Biowulf before you start the testing job:
  https://github.com/mxhulab/cryopros?tab=readme-ov-file

• Use system provided option (--opt $TRAIN_OPT) or user customized json file for training.
• Please note the particle training step allow the utilization of multiple GPUs, for example if allocating 2 GPUs for interactive or sbatch job, it needs to be specified in the script/command: --gpu_ids 0 1
• To generate auxiliary particles, please only allocate and use single GPU that specified as an environment: export CUDA_VISIBLE_DEVICES=0

Allocate an interactive session and run the program. Sample session for cryopros-train and cryopros-generate (user input in bold, check tutorial for other steps):

[user@biowulf]$ sinteractive --gres=gpu:v100x:2,lscratch:100 --cpus-per-task=8 --mem=40g
salloc: Pending job allocation 5879778
salloc: job 5879778 queued and waiting for resources
salloc: job 5879778 has been allocated resources
salloc: Granted job allocation 5879778
salloc: Waiting for resource configuration
salloc: Nodes cn4466 are ready for job

[user@cn4466 ~]$ module load cryopros
[user@cn4466 ~]$ cryopros-train --help
[user@cn4466 ~]$ cd /lscratch/$SLURM_JOBID
[user@cn4466 ~]$ cp -r $CRYOPROS_EXAMPLES ./
[user@cn4466 ~]$ cd examples
[user@cn4466 examples]$ cryopros-train \
> --opt $TRAIN_OPT \
> --gpu_ids 0 2 \
> --task_name HAtrimer_iteration_1 \
> --box_size 256 \
> --Apix 1.31 \
> --volume_scale 50 \
> --init_volume_path 6idd_align_lp10.mrc \
> --data_path T00_HA_130K-Equalized-Particle-Stack.mrcs \
> --param_path autorefinement.star \
> --invert \
> --dataloader_batch_size 8 \ 
> --dataloader_num_workers 0

[user@cn4466 examples]$ export CUDA_VISIBLE_DEVICES=0
[user@cn4466 examples]$ cryopros-generate \
> --model_path HAtrimer_iteration_1.pth \
> --param_path autorefinement.star \
> --output_path generated_HAtrimer_iteration_1 \
> --gen_name HAtrimer_iteration_1_generated_particles \
> --batch_size 50 \
> --box_size 256 \
> --Apix 1.31 \
> --invert \
> --gen_mode 2 \

cryopros-train sample session (user input in bold):

[user@biowulf]$ cat << 'EOS' > cryopros_train.sh 
> #!/bin/bash
> ml cryopros
> cryopros-train \
> --opt $TRAIN_OPT \
> --gpu_ids 0 2 \
> --task_name HAtrimer_iteration_1 \
> --box_size 256 \
> --Apix 1.31 \
> --volume_scale 50 \
> --init_volume_path 6idd_align_lp10.mrc \ 
> --data_path T00_HA_130K-Equalized-Particle-Stack.mrcs \ 
> --param_path autorefinement.star \ 
> --invert \ 
> --dataloader_batch_size 8 \ 
> --dataloader_num_workers 0 
> EOS
[user@biowulf]$ sbatch --partition=gpu --gres=gpu:v100x:2 --cpus-per-task=8 --mem=40g cryopros_train.sh

cryopros-generate sample session (user input in bold):

[user@biowulf]$ cat << 'EOS' > cryopros_generate.sh 
> export CUDA_VISIBLE_DEVICES=0
> cryopros-generate \
> --model_path HAtrimer_iteration_1.pth \
> --param_path autorefinement.star \
> --output_path generated_HAtrimer_iteration_1 \
> --gen_name HAtrimer_iteration_1_generated_particles \
> --batch_size 50 \
> --box_size 256 \
> --Apix 1.31 \
> --invert \
> --gen_mode 2
> EOS
[user@biowulf]$ sbatch --partition=gpu --gres=gpu:v100x:1 --cpus-per-task=8 --mem=30g cryopros_generate.sh