Denoising Tomograms in WarpTools

WarpTools comes with everything you need to denoise tomograms using the Noise2Noise denoising scheme.

Outline

Tomogram denoising is performed in two steps 1. generate tomograms from half sets of data 2. train and apply a denoiser using Noise2Map

Noise2Noise denoising requires two independent noisy observations of a target signal. In cryo-EM we can generate independent noisy copies of the same image by averaging the even and odd frames in a frame series.

We perform reconstructions from these even and odd images to produce independently noisy tomograms for denoising.

Generating tomograms from half sets of data

Generating even and odd frame series averages

We can generate even and odd movie averages by adding the --out_average_halves flag when running WarpTools fs_motion_and_ctf. If you didn't do this don't worry, you can run WarpTools fs_export_micrographs with the --average_halves flag.

This will produce even and odd subfolders filled with images in the average folder of your frame series processing directory.

Generating even and odd tomograms

Now that we have even and odd images, producing even and odd tomograms is as simple as running WarpTools ts_reconstruct with the --halfmap_frames flag.

What if I don't have frame series?

Don't worry! You can generate tomograms from even and odd tilts instead. Just add --halfmap_tilts to your ts_reconstruct command.

This will generate even and odd subfolders filled with images in the reconstruction folder of your tilt series processing directory.

Training and applying the denoiser

Now we can use Noise2Map to train a denoiser and apply it to our tomograms. We typically need to train for >10,000 iterations to get good results.

An example Noise2Map command for tomogram denoising is provided below.

Noise2Map
--observation1 warp_tiltseries/reconstruction/even \
--observation2 warp_tiltseries/reconstruction/odd \
--observation_combined warp_tiltseries/reconstruction \
--dont_flatten_spectrum \
--dont_augment