ifcoder 0.1.0

IFcoder: embeddings of immunofluorescence cell images

IFcoder

IFcoder is a Python package for embedding immunofluorescence cell images into the latent space using variational autoencoders (VAEs).

It is designed to operate seamlessly with CellProfiler and Scanpy, making it easy to move from image segmentation to downstream single-cell analysis.

---

Key features

• Extracts image patches centered at cell objects based on CellProfiler segmentation.
• Learns embeddings of the image patches using a convolutional VAE.
• Stores the embeddings in a single AnnData file along with the image patches, CellProfiler measurements, and metadata.

---

Installation

pip install ifcoder

Input

• CellProfiler output CSV. Four columns Location_Center_X, Location_Center_Y, ImageNumber, ObjectNumber are mandatory. Optionally, user can append additional columns of metadata (e.g., batch, treatment, diagnosis).
• Input images of the CellProfiler pipeline. Image patches will be extracted from them.

Example

please see ifcoder_example.ipynb.

Usage

1. Extract image patches based on CellProfiler output

ifcoder extract --cp-csv ${cellprofiler_csv} --out ${patches_h5ad}

This step:

• reads a CellProfiler output (e.g. Cells.csv),
• extracts image patches centered on cells,
• saves the image patches, metadata, and measurements in an AnnData file

2. Train VAE and compute embeddings

ifcoder train --data ${patches_h5ad} --out ${embeddings_h5ad}

This runs VAE and adds learned embeddings to the AnnData object adata.obsm["X_ifcoder"]

3. Downstream analysis

import scanpy as sc

adata = sc.read_h5ad("embeddings.h5ad")
sc.pp.neighbors(adata, use_rep="X_ifcoder")
sc.tl.umap(adata)
sc.pl.umap(adata)

This example draws UMAP dimentionality reduction of cells.

Output

IFcoder uses AnnData as the central data structure:

• adata.obsm["X_ifcoder"]

  Learned VAE embeddings of cell objects.

• adata.obsm["patches"]

  Image patches of cell objects with dimensions (n_cells, n_channels, height, width).

• adata.X

  CellProfiler quantitative measurements of cell objects. The columns starting with AreaShape_ and Intensity_ in CellProfiler output are stored here.

• adata.obs

  Cell metadata (the columns without prefixes AreaShape_ and Intensity_).