AnnoSpat automatically annotates cell types and quantifies inter-cellular arrangements from spatial proteomics

Aanchal Mongia, Diane Sauders, Alvin C. Powers, Ali Naji, Gregory W. Schwartz and Robert B. Faryabi

Introduction

Annospat is a tool for annotating and inferring patterns in single cells from spatial proteomics. It uses neural network and point process algorithms to automatically identify cell types and quantify cell-cell spatial relationships in the absence of manual annotation. Using the canonical markers for each protein in the antibody panel , it predicts the cell type labels by rendering the similarity between the cells in the proteomic space.

Implementation

Annospat can be used as a python library or as a command line tool using the argument generateLabels which takes as input the path to the raw proteomics matrix (with cells on rows and proteins as columns) and a signature file holding the canonical protein markers with other data related inputs. The tool is available for installation at https://pypi.org/project/AnnoSpat/1.0.0/

Installation

Dependencies: Python 3.6+, pip Libraries->typer, numpy, scikit-learn, pandas

pip install AnnoSpat==1.0.0

From your working directory, execute:

mkdir outputdir

AnnoSpat generateLabels -i <path_to_proteomics_matrix> -m <path_to_marker_file> -o outputdir -f <first_protein_name> -l <last_protein_name> -r <name_of_col_holding_ROInames>

Please replace the arguments to --inputfile/-i argument, --markerfile/-m argument and other arguments as per your own paths to proteomics and marker files and data.

Usage: AnnoSpat generateLabels [OPTIONS]

  Generate cell type annotations

Options:
  -i, --inputfile TEXT            [required] #path ot input proteomics file
  -m, --markerfile TEXT           [required] #path to marker file
  -o, --outputdir TEXT            [required] #path to output dir
  -f, --firstprotein TEXT         [required] #first protein to pick in the proteomics file
  -l, --lastprotein TEXT          [required] #last protein to pick in the proteomics file
  -r, --roicol TEXT               [required] #ROI column
  -n, --colsToNeglect TEXT        [default: ] #proteins to negelct in the proteomics file
  -d, --diseasecol TEXT           [default: ] # T1D/control
  -s, --suffix TEXT               [default: _Data_AnnoSpat] #suffix to be given
  -c, --classifier TEXT           [default: ELM] #classifier to use 
  -t, --thresholdmaxandunknown TEXT
                                  [default: [99.999, 70]] #thresholds for each protein default: b/w 99 and 99.9
  -a, --thresholdadaptive TEXT    [default: [99.5,99.5,99.5,99.5,99.9,
                                  99,99.5,  99,99,  99.5,99.9,
                                  99.9,99.9,99.9,  99.5,99.5]] #adaptive thresholds for each protein

  -b, --fileseparator TEXT        [default: ,] #file spearator    

NOTE: The number of values in "adaptive_th_vec_str" is assumed to be the same as no of cell types as each value by definition correspond to the threshold to be taken for each cell type.

Once the cell type annotations are obtained, The neighborhood analysis with point process can be implemented as shown in as described at: Too-Many-Cells