2D particle picking

nextPYP implements four methods for particle picking within the Pre-processing block

Particle positions are stored as “lists”. Multiple lists can be saved for each Pre-processing block

Import coordinates

External coordinates can be imported as EMAN box files (*.box) or IMOD model files (*.spk):

  1. Open the settings of the Pre-processing block and go to the Particle detection tab

  2. Select “import” as the particle picking Method

  3. Specify the particle radius in A (for visualization purposes)

  4. Select the location to Import particle coordinates (*.box, *.spk) (the folder should contain separate .box or .spk files for each micrograph)

  5. Click Save,, Run, and Start Run for 1 block to update the block

  6. Navigate to the Pre-processing block and go to the Micrographs tab to confirm that the particles were imported correctly

Manual picking

nextPYP provides a convenient UI to pick particles manually:

  1. Navigate to the Pre-processing block and go to the Micrographs tab

  2. Create a new list of particles by entering a name and clicking New

  3. Select particles in the current micrograph by simply clicking on them

  4. Navigate to other micrographs in the dataset using the navigation bar and select more particles as needed

  5. Open the settings of the Pre-processing block and go to the Particle detection tab

  6. Select “manual” as the particle picking Method

  7. Specify the particle radius in A (for visualization purposes)

  8. Choose the list of manually selected positions from the Select list for training dropdown menu

  9. Click Save, Run, and Start Run for 1 block to update the block

Tips

  • Particles can be deleted by right-clicking on the green markers

  • Particle positions are saved automatically every time a particle is added or deleted

  • The total number of particles in a dataset is displayed at the top of the page

  • It is possible to Copy, Delete, and Load lists

Size-based picking

This method selects particles based on their size:

  1. Open the settings of the Pre-processing block and go to the Particle detection tab

  2. Select “auto” or “all” as the particle picking Method (“auto” is more conservative, “all” tends to overpick)

  3. Specify the particle radius in A and other parameters as needed

  4. Click Save, Run, and Start Run for 1 block to update the block

  5. Navigate to the Pre-processing block and go to the Micrographs tab to inspect the results

Neural-network picking

Neural-network based methods require an existing list of particles for training a model. To pick particles, the trained model is then evaluated on the entire dataset. nextPYP uses a self-supervised approach that only needs sparsely annotated data. A wrapper for Topaz picking is also included.

Training

  1. Open the settings of the Pre-processing block and go to the Particle detection tab

  2. Select “pyp-train” or “topaz-train” as the particle picking Method

  3. Go to the corresponding Training/Evaluation tab and set the desired parameters

  4. Choose a list of positions from the Select list for training dropdown menu

  5. Click Save, Run, and Start Run for 1 block to train the model

Tips

  • Since training runs on the GPU, a standalone GPU-server is required (or GPU partitions must be properly configured in SLURM)

  • The trained model(s) are saved in the project directory under: train/YYYYMMDD_HHMMSS/*.training

  • Challenging datasets may require the use of more particles for training

Evaluation

  1. Open the settings of the Pre-processing block and go to the Particle detection tab

  2. Select “pyp-eval” or “topaz-eval” as the particle picking Method (depending on which method was used for training)

  3. Go to the corresponding Training/Evaluation tab and specify the location of the trained model (*.training file)

  4. Click Save, Run, and Start Run for 1 block to pick particles on all micrographs

  5. Navigate to the Pre-processing block and go to the Micrographs tab to inspect the results