nuc2seg

Welcome to the documentation for the Python implementation of nuc2seg

nuc2seg is method for cell body segmentation of 10X Xenium data.

The default Xenium analysis includes very accurate nucleus segmentation via DAPI staining, but there is no comparable cell body segmentation included. The cell body segmentation that is provided uses a simple expansion of the nuclear segments that is not always accurate ( https://kb.10xgenomics.com/hc/en-us/articles/11301491138317-How-does-Xenium-perform-cell-segmentation ). nuc2seg solves the problem of cell body segmentation by using information about the distribution of transcripts and the gene expression profiles of cell types in the slide to determine better cell body segments.

nuc2seg is provided as an nf-core compatible nextflow pipeline. All standard nf-core pipeline features are available. Read more about nf-core here: https://nf-co.re/docs/usage/introduction

Quickstart

nextflow run tansey-lab/nuc2seg \
    -r main \
    -profile <docker/singularity/iris/...> \
    --xenium_dir <path to xenium output> \
    --wandb_api_key <optional weights and bias api key for tracking UNet training> \
    --outdir /your/outdir \
    -w /your/outdir/nf

If running on MSKCC iris cluster, see the Running on MSKCC Iris Cluster section for instructions.

The nextflow pipeline also provides the following optional parameters:

  • --dataset: Existing preprocessed.h5 file to use, will skip preprocessing.

  • --celltyping_results: Existing celltyping results to use (i.e. --celltyping_results "/your/outdir/cell_typing_chain_*.h5").

  • --weights: Existing .ckpt to use, will skip neural net training.

  • --resume_weights: Existing .ckpt to use, will initialize model weights to this checkpoint and continue training.

  • --sample_area: Area of the slide to clip in bounding box format x1,y1,x2,y2 (e.g. --sample_area "0,0,1000,1000").

Contents

Indices and tables