Add Annotation Importer for Instance segmentations.

[uermel]

Imports TARDIS instance points, new point formate with the additional label "instance_id".

[uermel]

A config file to test with current ingestion tools:

dataset:
  dataset_identifier: 10002
  dataset_description: "Cryo-electron tomograms of RPE1 cells. Comprehensive annotation of actin filaments and microtubules"
  dataset_title: RPE1 cytosol with actin stress fiber
  authors: &dataset_authors
    - name: Irene de Teresa Trueba
      ORCID: 0000-0002-4691-9501
      primary_author_status: true
    - name: Sara Goetz
      ORCID: 0000-0002-9903-3667
    - name: Alexander Mattausch
      ORCID: 0000-0003-0901-8701
    - name: Frosina Stojanovska
      ORCID: 0000-0002-4327-1068
    - name: Christian Eugen Zimmerli
      ORCID: 0000-0003-4388-1349
    - name: Mauricio Toro-Nahuelpan
      ORCID: 0000-0001-5333-3640
    - name: Dorothy W. C. Cheng
    - name: Fergus Tollervey
    - name: Constantin Pape
      ORCID: 0000-0001-6562-7187
    - name: Martin Beck
      ORCID: 0000-0002-7397-1321
    - name: Alba Diz-Munoz
      ORCID: 0000-0001-6864-8901
    - name: Anna Kreshuk
      ORCID: 0000-0003-1334-6388
    - name: Julia Mahamid
      ORCID: 0000-0001-6968-041X
      corresponding_author_status: true
    - name: Judith B. Zaugg
      ORCID: 0000-0001-8324-4040
      corresponding_author_status: true
  cross_references:
    dataset_publications: &publications doi:10.1101/2022.04.12.488077, doi:10.1038/s41592-022-01746-2
    related_database_entries: EMPIAR-10989
  cell_component:
    name: ~
    id: ~
  cell_type:
    id: CL:0002586
    name: Retinal pigment epithelial-1
  dates: &repo-dates
    deposition_date: 2023-05-01
    last_modified_date: 2023-11-08
    release_date: 2023-11-30
  funding:
    - funding_agency_name: European Research Council (ERC)
      grant_id: '760067'
  grid_preparation: 'model: Quantifoil, material: GOLD, support_film_film_type_id: 1, support_film_film_topology: HOLEY, support_film_instance_type: support_film, pretreatment_type_: PLASMA CLEANING'
  key_photos:
    snapshot: https://www.ebi.ac.uk/pdbe/emdb-empiar/entryIcons/10989-l.gif
    thumbnail: https://www.ebi.ac.uk/pdbe/emdb-empiar/entryIcons/10989.gif
  organism:
    name: Homo sapiens
    taxonomy_id: 9606
  sample_preparation: 'buffer_ph: 7.4, vitrification_cryogen_name: ETHANE, cryo_protectant: None, instance_type: tomography_preparation'
  sample_type: cell
  tissue:
    id: BTO:0001175
    name: retina
runs: {}
tiltseries:
  acceleration_voltage: 300000
  binning_from_frames: 1
  camera:
    manufacturer: Gatan
    model: K2 SUMMIT
  data_acquisition_software: SerialEM
  microscope:
    manufacturer: TFS
    model: KRIOS
  microscope_optical_setup:
    energy_filter: GIF Quantum LS
    phase_plate: VOLTA PHASE PLATE
    image_corrector: null
  pixel_spacing: '{run_pixel_spacing}'
  related_empiar_entry: ~
  scales: []
  spherical_aberration_constant: 2.7
  tilting_scheme: Dose symmetric from 0.0 degrees
  tilt_axis: 79
  tilt_range:
    min: -60
    max: 60
  tilt_step: 3
  tilt_series_quality: 5
  total_flux: 125
  is_aligned: false
  alignment_binning_factor: ~ # We may need to calculate this for each tilt if we can identify aligned & unaligned files in the bucket.
tomograms:
  ctf_corrected: false
  fiducial_alignment_status: NON_FIDUCIAL
  offset:
    x: 0
    y: 0
    z: 0
  affine_transformation_matrix: [[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, -1, 0], [0, 0, 625, 1]] # TODO - this is different per tomo
  processing: raw
  reconstruction_method: Weighted back projection
  reconstruction_software: IMOD
  tomogram_version: 1
  authors: *dataset_authors
  voxel_spacing: '{run_voxel_spacing}'
annotations:
  - metadata:
      annotation_object:
        id: GO:0016020
        name: membrane
        description: ~
        state: ~
      dates: *repo-dates
      annotation_method: TARDIS
      method_type: automated
      annotation_publications: ~
      ground_truth_status: False
      authors: &annotation_authors
        - name: Robert Kiewisz
          primary_annotator_status: true
        - name: Tristan Bepler
          corresponding_author_status: true
      annotation_software: TARDIS
      version: '1.0'
      confidence:
        precision: ~
        recall: ~
      is_curator_recommended: False
    sources:
      - file_format: tardis
        binning: 1
        order: xyz
        shape: InstanceSegmentation
        glob_string: 10002/{run_name}/{run_name}_instance.csv
        is_visualization_default: false
  - metadata:
      dates: *repo-dates
      annotation_method: TARDIS
      method_type: automated
      annotation_publications: *publications
      ground_truth_status: False
      authors: *annotation_authors
      annotation_software: TARDIS
      version: '1.0'
      confidence:
        precision: ~
        recall: ~
        ground_truth_used: ~
      annotation_object:
        id: GO:0016020
        name: membrane
        description: ~
        state: ~
      is_curator_recommended: false
    sources:
      - file_format: mrc
        shape: SemanticSegmentationMask
        glob_string: 10002/{run_name}/{run_name}_semantic.mrc
        is_visualization_default: false

overrides_by_run:
  - run_regex: "^00011$"
    tiltseries:
      pixel_spacing: 3.45
    tomograms:
      voxel_spacing: 13.8
  - run_regex: "^00012$"
    tiltseries:
      pixel_spacing: 3.45
    tomograms:
      voxel_spacing: 13.8
standardization_config:
  destination_prefix: '10002_robert'
  source_prefix: julia_test/RPE1/RPE1
  run_to_frame_map_csv: run_to_frame_name_map.csv
  gain_glob: CountRef_{mapped_frame_name}-range.dm4
  frames_glob: frames/{mapped_frame_name}_*.tif
  rawtlt_files:
    - metadata/{run_name}_sq_df_sorted_fid.xf
    - metadata/{run_name}.mdoc
  tiltseries_glob: stack/{run_name}_sq_df_sorted_orig.st
  tomo_format: mrc
  tomo_glob: tomograms/{run_name}_*.rec
  tomo_voxel_size: ''
  run_glob: tomograms/*.rec
  run_regex: .*
  run_name_regex: (.*)_sq_df_sorted.rec
  run_data_map_file: run_data_map.csv

[uermel]

@jgadling

In creating this for the short term I have some thoughts on how we import point annotations:

• ideally the voxel_spacing should be available to the Importer.load method so point coordinates can be scaled by the voxel_size should they be given in Angstrom
• we should consolidate all point importers and maybe follow the pattern I suggest in this PR (a dataclass defining the point format)