default-config.yaml

###
# Config file for the STARCall processing pipeline.
# make a copy of this file and change any relevant parameters.
###

# Scale of objective used to image phenotyping and genotyping images.
# The absolute value of each is not important, but the relative ratio
# is used to scale up and down images. Both should be integers.
phenotype_scale: 20
bases_scale: 10

# The channels imaged for sequencing and phenotyping images.
# Channels for each base should be labeled as such so bases
# are called properly. 
sequencing_channels: ['DAPI', 'GFP', 'G', 'T', 'A', 'C']
phenotyping_channels: ['DAPI', 'GFP', 'Ph+WGA', 'Mito']

# Grid size for different processing steps. Input images are split
# into an X by X grid, the processing is run, then resulting images are
# merged back together.
# This can be disabled by setting the size to 1 or not setting the variable
segmentation_grid_size: 4
sequencing_grid_size: 4
phenotyping_grid_size: 12
# Segmentation grid sizes can be further specified, this is really only useful if
# you are running segmentation on the gpu and need very fine grain control
#segmentation_cells_grid_size: 6
#segmentation_nuclei_grid_size: 3

stitching:
    # The channel used to align images, can be an index or the string
    # name of one of the channels above. If a string, it should be
    # present in both sequencing and phenotyping channels, to align between them.
    #channel: 0
    channel: 'GFP'
    # The amount of overlap in pixels between generated tiles. This should be
    # larger than a cell, so that all cells are fully contained in at least one tile
    overlap: 100
    # The degree to which the stitching algorithm will align to sub pixel precision,
    # so 16 would mean alignment is done to 1/16 of a pixel. This requires minimal extra
    # processing, and improves alignment up to 16, past which improvements are not small.
    # This can be disabled by not setting this or setting it to 1.
    subpixel_alignment: 16
    # Solving method for alignment. I have found 'mae' (Mean absolute error minimizing solver)
    # to be the best in most situations. Other options are 'mse' (Mean squared error minimizing solver),
    # 'spantree' (Spanning tree)
    solver: 'mae'
    # The limit on the number of pairwise alignments between cycles. Eg, if
    # the limit is 5, cycle 0 would be aligned to cycle 4 but not cycle 5.
    # This reduces the processing greatly when the number of cycles is large.
    max_cycle_pairs: 16
    # The method to combine overlapping tiles. Mean takes the mean, nearest crops
    # each tile to be not overlapping
    merger: 'nearest'

    # This section is all of the parameters for ASHLAR stitching. ASHLAR may provide
    # better stitching than ConStitch, but it should be noted that ASHLAR cannot be used
    # when the phenotype cycles are imaged at a different scale
    ashlar:
        flip_x: False
        flip_y: False
        transpose: False
        interp: false
        filter_sigma: null
        maximum_shift: 1000
    ashlar_executable: 'ashlar'
    ashlar_overlap: 0.15

segmentation:
    # Diameter passed to cellpose, in pixels of the phenotype images.
    diameter: 100

    # Method to segment nuclei. Can be 'otsu_threshold' or 'stardist'
    nuclei_method: 'stardist'
    # Method to segment cells. Can be 'cellpose' or 'stardist'
    cells_method: 'cellpose'

    # Channels used to run cellpose, can be indices or strings from the
    # list of phenotype channels above. The first channel should be the nuclear
    # channel and the second the cytoplasm channel
    channels: ['DAPI', 'Ph+WGA']

    # Whether the cell and nuclei masks should be linked together.
    # Only necessary if you are going to use the nuclei masks, for example
    # in phenotyping
    match_masks: True

    # Whether cell segmentation is run on background corrected images
    # if both this and phenotyping.use_corrected are false, background correction
    # will not be run
    use_corrected: False

dotdetection:
    # Parameters for the LoG blob detection algorithm that detects dots.
    # These values should work well for most images, but if dot detection is
    # not working well, they can be adjusted
    min_sigma: 1
    max_sigma: 3
    num_sigma: 7

read_clustering:
    # Parameters for clustering nearby reads into single reads of the same sequence
    # Distances between reads are calculated with three values, positional distance, 
    # cosine distance between their read values extracted from the sequencing images,
    # and edit distance between their sequences.
    # The current values for the paramters (99999, 0, 1 for weights, linkage max and threshold 0.5)
    # simply combines all reads that have the same sequence. This system is still being
    # tested, and experimentation with parameters is encouraged.
    normalization: 'none'
    positional_weight: 99999
    value_weight: 0
    sequence_weight: 1

    # These distances are then clustered with agglomerative clustering, specified
    # with linkage (min, mean, max) and threshold. More on agglomerative clustering
    # can be found here
    # https://scikit-learn.org/stable/modules/generated/sklearn.cluster.AgglomerativeClustering.html#sklearn.cluster.AgglomerativeClustering
    linkage: 'max'
    threshold: 0.5

sequencing:
    # The max number of reads kept for each cell. A good number for this is usually,
    # 4-8, but this depends on how prevelant errors are and the size of the barcode library
    max_reads: 4

phenotyping:
    # Whether cell segmentation is run on background corrected images
    # if both this and sequencing.use_corrected are false, background correction
    # will not be run
    use_corrected: False

    # The executable to run to invoke cellprofiler. If cellprofiler is installed
    # in a different conda environment, this should reflect that.
    cellprofiler_executable: 'cellprofiler'
    #cellprofiler_executable: '~/miniconda3/envs/cp4/bin/cellprofiler'


# Directory that contains the raw .nd2 microscope files. It should
# contain one subdirectory for each cycle, each with one nd2 file for each well.
# If your files are not in nd2 format, they can be converted to tif files
# and placed in the input directory, described below.
# For example:
#   rawinput/
#   ├── 20231215_180258_893
#   │   ├── Well1_ChannelDAPI,GFP,G,T,A,C_Seq0000.nd2
#   │   │       ...
#   │   └── Well6_ChannelDAPI,GFP,G,T,A,C_Seq0003.nd2
#   │           ...
#   ├── 20231219_001028_540
#   │   ├── Well1_ChannelDAPI,GFP,G,T,A,C_Seq0000.nd2
#   │   │       ...
#   │   └── Well6_ChannelDAPI,GFP,G,T,A,C_Seq0003.nd2
#   └── phenotype
#       ├── Well1_Channel408 nm,473 nm,545 nm,635 nm_Seq0000.nd2
#       │       ...
#       └── Well6_Channel408 nm,473 nm,545 nm,635 nm_Seq0003.nd2
rawinput_dir: 'rawinput/'

# Directory that contains the input in .tif file form. If the data is
# present in the rawinput dir, this will be automatically filled.
# The directory should contain one subdirectory for each well, each
# with a subdirectory for each cycle, containing 'raw.tif' with the unstitched
# tile images and 'positions.csv' with the positions of each tile.
# An additional subdirectory 'auxdata' can be provided containing a barcode
# library.
# For example:
#    input/
#    ├── auxdata
#    │   └── barcodes.csv
#    └── well1
#        ├── cycle00
#        │   ├── positions.csv
#        │   └── raw.tif
#        │      ...
#        ├── cycle11
#        │   ├── positions.csv
#        │   └── raw.tif
#        └── cyclePT
#            ├── positions.csv
#            └── raw.tif
#       ...
input_dir: 'input/'

# Directory that contains all files generated when stitching unstitched
# tiled image sets. If your input images are already stitched, you can
# place them in this directory, following the pattern 'well{well}/raw.tif',
# where the tiff file is of shape (num_cycles, num_channels, width, height)
stitching_dir: 'stitching/'

# Directory containing all files generated when performing cell segmentation
segmentation_dir: 'segmentation/'

# Directory that contains all files generated when sequencing the in situ
# sequencing reads
sequencing_dir: 'sequencing/'

# Directory that contains all files generated when calculating features and
# phenotyping cells
phenotyping_dir: 'phenotyping/'

# Folder for final output
output_dir: 'output/'


# If necessary the individual wells and cycles can be specified here, otherwise
# they will be automatically determined by the folders in rawinput/ or input/.
#wells: ['B2', 'C2']
#cycles: ['00', '01', '02', '03', '04', '05', '06', '07']

# The prefix for phenotyping cycles in rawinput/, all directories that begin
# with this string such as phenotype/, phenotype2/, phenotype_20250225/
# are all considered phenotyping cycles.
phenotype_date: 'phenotype'
# The cycle name for each phenotyping cycle, to be stored in input/ and such.
#phenotype_cycles: ['PT']