Python package and desktop application · BSD-3-Clause

spaCR

Spatial phenotype analysis of CRISPR-Cas9 screens.

spaCR segments cells, nuclei and pathogens in microscopy images, measures every object, and links the measurements to the guide RNAs sequenced from the same wells. It starts from plate images and FASTQ reads and ends with per-cell measurements, trained classifiers and a ranked list of genes.

Version 1.5.1.4
Released 2026-10-09
Python 3.9–3.14

Install Documentation Tutorials Source code

Hoechst image of infected HeLa cells with nuclei outlined in blue, host cells in cyan and Toxoplasma vacuoles in magenta — One Hoechst image, three object classes: nuclei (blue), host cells (cyan) and Toxoplasma vacuoles (magenta). Host cells and vacuoles are predicted from the Hoechst channel alone. See Models.

Pipeline

One field followed through spaCR: plate 1, well H02, field 4 of the example screen, a well whose guides are 92% GRA14, the positive control.

1Image
19 infected cells, reassembled from Measure crops.
2Mask
Cell, nucleus and parasite outlines.
3Measure
Cells coloured by parasite channel-1 intensity.
4Classify
Classifier score per cell; 18 of 19 above 0.5.
5Hit
GRA14 called by regression (q 0.040).

Data: spacr-example-screen crops and spacr-example-hit measurements, scores and guide counts. Cell outlines are the screen's masks; nucleus and parasite outlines were recomputed with Cellpose-SAM through spaCR on CPU; the regression covers this 30-well subset.

MaskSegments cells, nuclei, pathogens and organelles with Cellpose models.
MeasureShape, intensity, texture and position of every object, and a crop of each, in SQLite.
AnnotateLabels object crops in a grid with keyboard shortcuts.
ClassifyTrains image or measurement classifiers and stores held-out performance with each checkpoint.
Map barcodesAssigns sequencing reads to wells and guide RNAs.
RegressionEstimates guide, gene and condition effects with intervals and a ranked hit list.

Published models

Each figure shows one real image and the objects that spaCR's published models detect in it. Drag the divider to compare the input with the outlines, and use the check boxes to show or hide a class. Counts are objects in the image shown. Weights resolved through spaCR's model zoo (nightly, 2026-10-10) and run with Cellpose 4.2.1.1 on CPU.

16published models
4cross-channel models
7object classes shown

Hoechst fluorescence image of HeLa cells with bright nuclei and faint cytoplasm — Hoechst image: nuclei, host cells and Toxoplasma

One Hoechst image of HeLa cells infected with Toxoplasma gondii. Host cells and parasite vacuoles are predicted from the Hoechst channel alone by two spaCR cross-channel models. Nuclei come from stock Cellpose-SAM on the same channel.

Nuclei 53 Stock Cellpose-SAM (cpsam_v2), same channel. Pixels below 500 set to 0, diameter 90 px

Host cells 61 cell_from_hoechst_v1. Test F1 0.870 on 451 fields

Toxoplasma vacuoles 63 toxoplasma_from_hoechst_v1. Test F1 0.569 on 463 fields

Input
Hoechst (nuclear stain)

Image
HeLa field, plate 1 well L08 field 12, test split (held out by well)

Licence
CC BY 4.0

Size
2000 × 2000 px, shown at 1024 px

Figure
With caption and checksums

Hoechst image: nuclei, host cells and Toxoplasma

One Hoechst image of HeLa cells infected with Toxoplasma gondii. Host cells and parasite vacuoles are predicted from the Hoechst channel alone by two spaCR cross-channel models. Nuclei come from stock Cellpose-SAM on the same channel.

Nuclei 53 Stock Cellpose-SAM (cpsam_v2), same channel. Pixels below 500 set to 0, diameter 90 px

Host cells 61 cell_from_hoechst_v1. Test F1 0.870 on 451 fields

Toxoplasma vacuoles 63 toxoplasma_from_hoechst_v1. Test F1 0.569 on 463 fields

Input
Hoechst (nuclear stain)

Image
HeLa field, plate 1 well L08 field 12, test split (held out by well)

Licence
CC BY 4.0

Size
2000 × 2000 px, shown at 1024 px

Figure
With caption and checksums

Whole-cell stain fluorescence image of the same HeLa field — Cell-mask image: Toxoplasma, nuclei and host cells

The same field imaged with a whole-cell stain. Parasite vacuoles and nuclei are predicted from this channel alone by two spaCR cross-channel models, so no nuclear or parasite stain is needed. Host cells come from stock Cellpose-SAM on the same channel.

Toxoplasma vacuoles 43 toxoplasma_from_cellmask_v1. Test F1 0.606 on 463 fields

Nuclei 56 nuclei_from_cellmask_v1. Test F1 0.888 on 453 fields

Host cells 80 Stock Cellpose-SAM (cpsam_v2), same channel. Default settings

Input
Cell mask (whole-cell stain)

Image
HeLa field, plate 1 well L08 field 12, test split (held out by well)

Licence
CC BY 4.0

Size
2000 × 2000 px, shown at 1024 px

Figure
With caption and checksums

Cell-mask image: Toxoplasma, nuclei and host cells

The same field imaged with a whole-cell stain. Parasite vacuoles and nuclei are predicted from this channel alone by two spaCR cross-channel models, so no nuclear or parasite stain is needed. Host cells come from stock Cellpose-SAM on the same channel.

Toxoplasma vacuoles 43 toxoplasma_from_cellmask_v1. Test F1 0.606 on 463 fields

Nuclei 56 nuclei_from_cellmask_v1. Test F1 0.888 on 453 fields

Host cells 80 Stock Cellpose-SAM (cpsam_v2), same channel. Default settings

Input
Cell mask (whole-cell stain)

Image
HeLa field, plate 1 well L08 field 12, test split (held out by well)

Licence
CC BY 4.0

Size
2000 × 2000 px, shown at 1024 px

Figure
With caption and checksums

Greyscale image of a crystal-violet stained well with pale plaques in a dark monolayer — Plaque assay: Toxoplasma plaques

A crystal-violet plaque-assay well from a published figure. Plaques are segmented by spaCR's plaque model. This well was not used to train the model.

Plaques 99 toxoplasma_plaque_v3. Candidate model, not yet the default. 98 plaques in the hand-drawn mask; F1 0.964 at IoU 0.5 on this well

Input
Crystal violet, greyscale

Image
Cabral G, et al. Microbiol Spectrum 2024, Fig. 2

Dataset
toxoplasma-plaque-dataset (held out from v3 training)

Licence
CC BY 4.0

Figure
With caption and checksums

Plaque assay: Toxoplasma plaques

A crystal-violet plaque-assay well from a published figure. Plaques are segmented by spaCR's plaque model. This well was not used to train the model.

Plaques 99 toxoplasma_plaque_v3. Candidate model, not yet the default. 98 plaques in the hand-drawn mask; F1 0.964 at IoU 0.5 on this well

Input
Crystal violet, greyscale

Image
Cabral G, et al. Microbiol Spectrum 2024, Fig. 2

Dataset
toxoplasma-plaque-dataset (held out from v3 training)

Licence
CC BY 4.0

Figure
With caption and checksums

All models, with model cards, metrics and training data: model zoo documentation and Hugging Face.

Screen results

A pooled CRISPR screen in Toxoplasma gondii for parasite genes that affect recruitment of the host ESCRT protein GFP-TSG101 to the parasite vacuole, analysed with spaCR. From the spaCR preprint.

4screen plates
2genes past 3σ with both classifiers
p 8.4 × 10⁻⁵GRA14 vs SAG1 recruitment

Two volcano plots of gene effect against significance; EAF1 and GRA14 lie above the threshold in both — Gene effects from a measurement classifier (XGBoost, left) and an image classifier (MaxViT, right); EAF1 and GRA14 pass 3σ in both. Preprint Fig. 5.

Volcano plot with SAG1 and GRA14 above the significance line — Regression on one plate (30 wells): the controls GRA14 (q 0.040) and SAG1 (q 0.006) pass q < 0.05. Data: spacr-example-hit.

Interactive volcano plot in the Regression module with one gene selected and its gene card below — Regression module: clicking a point opens the gene card, here EAF1 (TGGT1_225160).

Grid of cell crops from a well that carries EAF1 guides — Cells from an EAF1 well whose classifier scores match the estimated effect.

Line plot of mean phenotype against gene rank with selected genes labelled — All genes ranked by mean phenotype; the strongest hits are labelled by gene ID.

Heat maps of unique guide RNAs per well on four plates — Map barcodes: unique guide RNAs per well on the four screen plates.

Phenotypes

Every measured cell can be placed on a two-dimensional map with its image crop, so that groups of similar cells can be found and checked by eye.

5,000cells mapped
5clusters
9%unassigned

UMAP of 5,000 cells with five outlined clusters and four example crops per cluster — Image UMAP of 5,000 cells from screen plate 1: five HDBSCAN clusters that differ in nuclei and vacuoles per cell; crops are cells near each cluster centre.

Measure module showing a grid of single-cell crops — Object crops saved by Measure from a sixteen-field run.

One field shown as separate channels followed by the cell, nucleus and parasite masks — One recruitment-assay field: channels, then cell, nucleus and parasite masks.

Assays and QC

Outputs of spaCR's assay and quality-control modules on its public example data sets, computed on CPU.

1,536wells, dose-response
24plates, control chart
99plaques in one well

Dose-response curves of median cell count for five compounds, SNX-2112 highlighted with its EC50 — Dose-Response: four-parameter logistic fits for the 5 of 58 compounds with a bounded EC50; SNX-2112 EC50 0.299 µM (95% CI 0.279–0.319). Data: spacr-example-dose, 1,536 wells, 4 plates.

Control chart of DMSO-well median cell count across 24 plates with control limits and flagged plates — Control chart: DMSO-well median cell count across 24 plates; all 24 fall outside the limits after cell-line, time-point and seeding changes. Data: spacr-example-control-chart.

Crystal-violet plaque well with 99 plaques outlined in magenta — Plaque Assay: 99 plaques found by toxoplasma_plaque_v3 in a well from Cabral et al. 2024 (CC BY 4.0); 98 in the hand-drawn mask.

Bar plots of GFP-TSG101 recruitment for negative and positive control wells on four plates — Recruitment: GFP-TSG101 vacuole-to-cytoplasm ratio in SAG1 (negative) and GRA14 (positive) control wells on four plates; plate means p = 8.4 × 10⁻⁵.

Papers using spaCR

Papers whose methods use spaCR, each checked in the full text; the quote under each entry is the evidence. Papers that only cite spaCR are not listed. Last checked 2026-10-10. To add a paper, open an issue.

A pooled image-based CRISPR screen identifies EAF1 as a T. gondii modulator of ESCRT subversion Olafsson EB, et al. bioRxiv, 2026, preprint. The spaCR paper
spaCR: Spatial phenotype analysis of CRISPR-Cas9 screens Olafsson EB, et al. Research Square, 2025, preprint. The spaCR paper Image: Fig. 3 from Olafsson et al. Research Square 2025, CC BY 4.0 (cropped)
Toxoplasma GRA8 engages the host ESCRT accessory protein ALG-2 and is necessary for parasite metabolic integrity Kaur H, et al. bioRxiv, 2026, preprint. Uses spaCR
All quantitative image analysis was performed as described in (33) in detail.Methods, ESCRT protein recruitment
Image: Fig. 4 from Kaur et al. bioRxiv 2026, CC0 1.0 (cropped)
IRE1α promotes phagosomal calcium flux to enhance macrophage fungicidal activity McFadden MJ, et al. Cell Reports, 2025. Uses spaCR. Preprint: bioRxiv 2023
For analysis of cellular calcium flux, the Python package spacr (https://github.com/EinarOlafsson/spacr) was used to segment and track cells over time and quantify single cell calcium oscillations.STAR Methods, 'Quantification of calcium flux'
Metabolic adaptability and nutrient scavenging in Toxoplasma gondii: insights from ingestion pathway-deficient mutants Rimple PA, et al. mSphere, 2025. Uses spaCR. Preprint: bioRxiv 2024
Plaques were detected and quantified using the plaque assay module in SpaCR (v. 0.3.62, https://github.com/EinarOlafsson/spacr and https://pypi.org/project/spacr/). First, plaques were manually …Methods, 'Validation of screens'
Image: Fig. S3 from Rimple et al. mSphere 2025, CC BY 4.0 (cropped)

Install spaCR 1.5.1.4

Desktop installers

The installers include their own Python, so Conda is not needed. Files from the v1.5.1.4 release.

Windows 10 and 11 spaCR-1.5.1.4-Windows-Online-Setup.exe Download 244 kB
macOS 11 or later, Intel and Apple silicon spaCR-1.5.1.4-macOS-Universal-Online.pkg Download 390 kB
Linux x86-64 spaCR-1.5.1.4-Linux-x86_64-Online.run Download 236 kB

Checksums: SHA256SUMS.txt. Older installers and troubleshooting: installer documentation.

PyPI

conda create -n spacr python=3.12 -y
conda activate spacr
python -m pip install spacr==1.5.1.4
spacr

conda-forge

conda create -n spacr python=3.12 -y
conda activate spacr
conda install conda-forge::spacr
spacr

spaCR on PyPI · spaCR on conda-forge · Requires Python !=3.14.1,<3.15,>=3.9

Cite spaCR

Olafsson EB, Arnold CS, Kellermeier JA, Rimple PA, Kaur H, Wang Y, Sexton JZ, Svärd S, Carruthers VB, O'Meara MJ. spaCR: spatial phenotype analysis of CRISPR-Cas9 screens (version 1.5.1.4). Zenodo; 2026. doi:10.5281/zenodo.23266544

All versions (concept DOI)	10.5281/zenodo.21343316
Version 1.5.1.4	10.5281/zenodo.23266544
Preprint	10.64898/2026.07.08.737057
Metadata	CITATION.cff

BibTeX

@software{olafsson_spacr_2026,
  author    = {Olafsson, Einar B. and Arnold, Christophe-Sebastien and Kellermeier, Jacob A. and Rimple, Patrick A. and Kaur, Hargobinder and Wang, Yifan and Sexton, Jonathan Z. and Svärd, Staffan and Carruthers, Vern B. and O'Meara, Matthew J.},
  title     = {{spaCR: spatial phenotype analysis of CRISPR-Cas9 screens}},
  version   = {1.5.1.4},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23266544},
  url       = {https://doi.org/10.5281/zenodo.23266544}
}

Documentation

The desktop application

Muted excerpts from the narrated tutorials. The full lessons, with narration and captions in several languages, are in the tutorial portal.

Home screen with one tile per module. Tutorial lesson 5.

Cellpose Workbench in Make Masks, showing the image, outlines and flows. Tutorial lesson 20.

spaCR

Pipeline

1Image

2Mask

3Measure

4Classify

5Hit