Tomogram denoising

nextPYP provides wrappers for several tomogram denoising methods based on neural networks, including cryoCARE, Topaz-Denoise and IsoNet

Denoising is done in two phases using the Denosing (training) and Denosing (eval) blocks. If a pre-trained model is available, the training phase can be skipped. Training is only supported for cryoCARE and IsoNet models (Topaz-denoise uses pre-trained models)

Training

  • Click on Tomograms (output of the Pre-processing block) and select Denoising (train)

  • Select the desired denoising method and corresponding parameters for training

  • (optional) To train models on a subset of tomograms, create a Filter in the Pre-processing block and select its name from the Filter tomograms dropdown menu

  • Click Save, Run, and Start Run for 1 block

  • Navigate to the Denoising (training) block to inspect the results of training

Note

This step can be skipped if a pre-trained model is available

Evaluation

  • Click on Denoising model (output of the Denoising (traiing) block) and select Denoising (eval)

  • Select the desired algorithm and corresponding trained model from the block upstream (cryoCARE and IsoNet) or list of pre-trained models (Topaz)

  • Click Save, Run, and Start Run for 1 block

  • Navigate to the Denoising (eval) block to inspect the denoised tomograms

Notes

  • Evaluation is typically done on the entire set of tomograms, while training is done using a subset of tomograms

  • cryoCARE and IsoNet need a GPU to run. Topaz can run on the CPU (default) or a GPU

Denoised tomograms can be used as input for all downstream operations available in nextPYP (e.g., particle picking, segmentation, etc.)

Note

All downstream particle refinement operations will still use the raw data (regardless of whether denoising was used for particle picking or not)