einarolafsson/spacr-example-hit

spaCR — Investigate Hit example

A small cut of the published spaCR GFP-TSG101 recruitment screen (einarolafsson/spacr-example-screen), with a Regression run made from it by spaCR. It is the data behind Load test data… on spaCR's Investigate Hit screen (an alpha feature): the screen opens the hit GRA14 (239740, the screen's positive control, whose loss removes TSG101 recruitment to the vacuole) with the guides the Regression names (239740_1), and traces it to candidate cells.

spacr-example-hit.tar: 27,791,360 bytes (27.8 MB), one uncompressed tar, SHA-256 da6fcbb5c282cc24080cd20a468c8fc334c898715e74b36179cd064245cbd2fd. guide_fractions.csv is the guide-fraction table, also loose here.

Source

HeLa GFP-TSG101 cells infected with a pooled Toxoplasma gondii CRISPR-Cas9 library, plate 1 of the four-plate screen:

• plate1-measurements.tar from einarolafsson/spacr-example-screen (spaCR Measure database: cell, cytoplasm, nucleus, pathogen and png_list tables);
• plate1_dv.csv (per-cell cross-validated predictions of spaCR's classifier: pred is the probability of the GRA14-knockout-like phenotype) and plate_1_unique_combinations.csv (guide read counts per well) from the spaCR example-data release.

Columns c1, c2 and c3 hold the controls: c1 is enriched for the negative control SAG1 (TGGT1_233460), c2 for the positive control GRA14 (TGGT1_239740), c3 a mix; the other columns carry the whole library.

What was cut

Fixed rules, no resampling, no edited values:

• 30 wells: columns c1, c2 and c3 of rows r1-r8, plus the 6 imaged wells whose reads hold no GRA14 guide at all (r16c7, r7c18, r14c19, r1c23, r3c23, r3c24);
• fields f1-f4 of the sixteen imaged per well;
• every row of every table of measurements.db for those fields (cell 2,608, nucleus 3,606, pathogen 5,456, cytoplasm 2,608, png_list 2,608), the matching 2,608 rows of plate1_dv.csv, and the 9,032 read-count rows of those wells (counts are per well, so the guide fractions are those of the whole well).

Made with spaCR on the cut

• results/ols/: spacr.ml.perform_regression (OLS, one row per well, mean pred, guide fraction threshold 0.02, control columns kept). It names 239740 (gene_fraction:gene[239740]: effect 0.482, p 0.00617, adjusted 0.0401).
• guide_fractions.csv: that run's regression_data.csv, one fraction per well and guide as the regression used them (a guide under 2 % of a well's reads counts as absent, so 16 wells carry GRA14 and 14 do not).
• hit.json: the hit and the file names, as the Load test data button fills them in.

A check run of spacr.hit_investigation.investigate_hit on this set cross-fitted 2,608 cells over 30 wells (held out by well). Hit-like prevalence in GRA14 wells minus GRA14-free wells was -0.0235: on this cut the morphology alone does not separate the two, which the screen reports honestly. Its outputs are not shipped.

This is a demonstration cut: 30 wells of one plate cannot stand for the screen, and the hit is the screen's built-in positive control. Use the full plates in spacr-example-screen for analysis.

Licence and citation

CC BY 4.0, as the source screen. When you use these data, please credit the TSG101 recruitment screen (einarolafsson/spacr-example-screen, Olafsson EB et al.) and cite spaCR:

• Olafsson EB, Arnold C-S, 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. Zenodo. doi:10.5281/zenodo.21343316

Built by tools/build_hit_example_dataset.py in spaCR.