config.vsh.yaml

name: popv
namespace: "annotate"
scope: "public"
description: "Performs popular major vote cell typing on single cell sequence data using multiple algorithms. Note that this is a one-shot version of PopV."
authors:
  - __merge__: /src/authors/matthias_beyens.yaml
    roles: [ author ]
  - __merge__: /src/authors/robrecht_cannoodt.yaml
    roles: [ author ]
argument_groups:
  - name: Inputs
    description: Arguments related to the input (aka query) dataset.
    arguments:
      - name: "--input"
        alternatives: [-i]
        type: file
        description: Input h5mu file.
        direction: input
        required: true
        example: input.h5mu
      - name: "--modality"
        description: Which modality to process.
        type: string
        default: "rna"
        required: false
      - name: "--input_layer"
        type: string
        description: Which layer to use. If no value is provided, the counts are assumed to be in the `.X` slot. Otherwise, count data is expected to be in `.layers[input_layer]`.
        required: false
      - name: "--input_obs_batch"
        type: string
        description: Key in obs field of input adata for batch information. If no value is provided, batch label is assumed to be unknown.
        required: false
      - name: "--input_var_subset"
        type: string
        description: Subset the input object with this column.
        required: false
      - name: "--input_obs_label"
        type: string
        description: Key in obs field of input adata for label information. This is only used for training scANVI. Unlabelled cells should be set to `"unknown_celltype_label"`.
        required: false
      - name: "--unknown_celltype_label"
        type: string
        description: If `input_obs_label` is specified, cells with this value will be treated as unknown and will be predicted by the model.
        default: "unknown"
        required: false
  - name: Reference
    description: Arguments related to the reference dataset.
    arguments:
      - name: "--reference"
        type: file
        description: "User-provided reference tissue. The data that will be used as reference to call cell types."
        example: TS_Bladder_filtered.h5ad
        direction: input
        required: true
      - name: "--reference_layer"
        type: string
        description: Which layer to use. If no value is provided, the counts are assumed to be in the `.X` slot. Otherwise, count data is expected to be in `.layers[reference_layer]`.
        required: false
      - name: "--reference_obs_label"
        type: string
        description: Key in obs field of reference AnnData with cell-type information.
        default: "cell_ontology_class"
        required: false
      - name: "--reference_obs_batch"
        type: string
        description: Key in obs field of input adata for batch information.
        default: "donor_assay"
        required: false
      # - name: "--reference_models"
      #   type: file
      #   description: Pretrained models. Can be a directory or a tar gz.
      #   required: false
      #   example: pretrained_models_Bladder_ts.tar.gz
  - name: Outputs
    description: Output arguments.
    arguments:
      - name: "--output"
        type: file
        description: Output h5mu file.
        direction: output
        required: true
        example: output.h5mu
    __merge__: [., /src/base/h5_compression_argument.yaml]

      # - name: "--output_models"
      #   type: file
      #   direction: output
      #   description: If `prediction_mode == "retrain"`, saves models to a directory and compresses the results into a tar gz.
      #   example: "output.tar.gz"
      #   required: false
  - name: Arguments
    description: Other arguments.
    arguments:
      - name: "--methods"
        type: string
        description: "Methods to call cell types. By default, runs to knn_on_scvi and scanvi."
        example: ["knn_on_scvi", "scanvi"]
        choices: [celltypist, knn_on_bbknn, knn_on_scanorama, knn_on_scvi, onclass, rf, scanvi, svm]
        required: true
        multiple: true
      # - name: "--prediction_mode"
      #   type: string
      #   description: |
      #     Execution mode of cell-type annotation.
      #     "retrain": Train all prediction models and saves them to disk. Argument `output_models` must be defined.
      #     "inference": Classify all cells based on pretrained models. Argument `reference_models` must be defined.
      #     "fast": Fast inference using only query cells and single epoch in scArches.
      # - name: "--plots"
      #   type: boolean
      #   description: "Creation of agreement and frequency plots between selected cell type algorithmn(s) and final PopV ensemble called cell type."
      #   default: false
      #   required: false
resources:
  - type: python_script
    path: script.py
  - path: /src/utils/setup_logger.py
  
test_resources:
  - type: python_script
    path: test.py
  - path: /resources_test/annotation_test_data/
  - path: /resources_test/pbmc_1k_protein_v3/

engines:
  - type: docker
    #image: nvcr.io/nvidia/pytorch:22.12-py3
    image: python:3.11-slim

    setup:
      - type: docker
        env:
          # Build extensions without AVX512/AVX2 support
          - CFLAGS="-mno-avx512f -mno-avx2" 
          - CPPFLAGS="-mno-avx512f -mno-avx2"
      - type: apt
        packages:
          - procps
          - git
          - build-essential
      - type: python
        packages:
          - popv~=0.4.2
          # Previously, scvi-tools < 1.2.2 pinned numpy to <2. We want to keep it pinned here
          - numpy<2
          - setuptools
      # These need to be updated AFTER popv is installed.
      # See https://github.com/YosefLab/PopV/issues/30
      - type: python
        __merge__: [ /src/base/requirements/anndata_mudata.yaml, .]
      # download ontology required by popv
      - type: docker
        run: |
          cd /opt && git clone --depth 1 https://github.com/YosefLab/PopV.git
    test_setup:
      - type: python
        __merge__: [ /src/base/requirements/viashpy.yaml, .]

runners:
  - type: executable
  - type: nextflow
    directives:
      # TODO: should add new label highmem-single-gpu and lowmem-single-gpu
      label: [highmem, highcpu, highdisk]