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
):
Open the settings of the Pre-processing block and go to the Particle detection tab
Select “import” as the particle picking
Method
Specify the particle radius in A (for visualization purposes)
Select the location to
Import particle coordinates (*.box, *.spk)
(the folder should contain separate.box
or.spk
files for each micrograph)Click Save,, Run, and Start Run for 1 block to update the block
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:
Navigate to the Pre-processing block and go to the Micrographs tab
Create a new list of particles by entering a name and clicking New
Select particles in the current micrograph by simply clicking on them
Navigate to other micrographs in the dataset using the navigation bar and select more particles as needed
Open the settings of the Pre-processing block and go to the Particle detection tab
Select “manual” as the particle picking
Method
Specify the particle radius in A (for visualization purposes)
Choose the list of manually selected positions from the
Select list for training
dropdown menuClick 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:
Open the settings of the Pre-processing block and go to the Particle detection tab
Select “auto” or “all” as the particle picking
Method
(“auto” is more conservative, “all” tends to overpick)Specify the particle radius in A and other parameters as needed
Click Save, Run, and Start Run for 1 block to update the block
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¶
Open the settings of the Pre-processing block and go to the Particle detection tab
Select “pyp-train” or “topaz-train” as the particle picking
Method
Go to the corresponding Training/Evaluation tab and set the desired parameters
Choose a list of positions from the
Select list for training
dropdown menuClick 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¶
Open the settings of the Pre-processing block and go to the Particle detection tab
Select “pyp-eval” or “topaz-eval” as the particle picking
Method
(depending on which method was used for training)Go to the corresponding Training/Evaluation tab and specify the location of the trained model (
*.training
file)Click Save, Run, and Start Run for 1 block to pick particles on all micrographs
Navigate to the Pre-processing block and go to the Micrographs tab to inspect the results