parameters-configuration-file.md

looptrace parameters configuration file

The parameters configuration file is where you tell looptrace about various algorithmic and data analysis settings to use, as well as where to read and write data.

Requirements and suggestions

• Check that each channel setting (often with a _ch or _channel suffix) matches what's been used in the imaging experiment which generated the data to be processed.
• xy_nm and z_nm should be adjusted to match your microscope settings (number of nanomerters per step in xy or in z).
• nuc_method should be set to "cellpose".
• nuc_3d should be absent or set to False.
• analysis_path should be an absolute path but may use environment and/or user variables.
• analysis_path should specify the path to a folder that exists before the pipeline is run.
• zarr_conversions should be a mapping from subfolder in the images folder (--images-folder when running from the command-line) to new subfolder (1-to-1): the keys are names of subfolders with raw image files (e.g., .nd2), and each value will be the new folder with that image data, just reformatted as .zarr. Typically there will be one entry for the sequential FISH images' folder and another for the nuclei images' folder.
• Check that spot_wavelength and objective_na have been adjusted to match the microscope and fluorophores used.
• decon_psf should be set to gen.
• require_gpu should be set to True.
• decon_iter should be set to a nonnegative integer. If you don't want to run deconvolution, set this to 0. Setting this to 0 obviates the need for NVIDIA and GPUs, so you could change nvidia-docker to simply docker when running the pipeline.
• decon_input_name should likely be the single value in the zarr_conversions mapping.
• reg_input_template and reg_input_moving should likely match each other and should correspond to adding a _decon suffix to the value of decon_input_name.
• reg_ref_timepoint should be set to something approximately equal to the middle of the imaging timecourse (i.e, midway between first and last timepoint).
• num_bead_rois_for_drift_correction should be set to 100 or 200 (number of beads to use for drift correction). Having bead count fewer than the value will impede processing, but higher values may given a bit better drift correction. Judge in accordance with how many beads you anticipate having per image.
• num_bead_rois_for_drift_correction_accuracy should be set to 100.
• coarse_drift_downsampling should be set to 2; use 1 for no downsampling.
• detection_method should be set to dog, and spot_threshold to 15. If using intensity, a much higher spot_threshold will be needed.
• subtract_crosstalk should be set to False, rendering crosstalk_ch irrelevant.
• pixelSeparationBeneathWhichSpotRoisWillMerge should be set to a positive value if you want to do mergers of regional barcode spot ROIs which are close together. If you don't wish to do that merger, omit this configuration key. The value will be interpreted as being in pixel (and $z$ slice) units, and represents a Euclidean distance.
• parallelise_spot_detection should be set to False.
• spot_downsample should be a small integer, often just 2 or even 1 (no downsampling).
• spot_in_nuc should be set to True, generally.
• padding_method should be set to edge.
• tracing_cores should be a value no more than the number of CPUs on the machine on which the processing will run.
• mask_fits should be set to False.
• roi_image_size should be a 3-element list and should most likely be (8, 16, 16) or (16, 32, 32).
• subtract_background should be set to 0.
• Check that the tracing QC parameters are as desired:
  • For A_to_BG, 2 is often a good setting.
  • For sigma_xy_max, 150 is often a good setting.
  • For sigma_z_max, 400 is often a good setting.
  • For max_dist, 800 is often a good setting.
• If you want the Numpy arrays representing the spot images for tracing (the *.npy files) to be kept even after zipping, set keep_spot_images_folder to True.