torchcell.sga

A colony-fitness pipeline adapted from SGAtools for single CRISPR knockouts dispensed onto plates by an ECHO acoustic liquid handler. It reads the gitter colony-size file and the ECHO picklist (io), corrects positional plate artifacts (normalize), and scores each knockout as fitness relative to the on-plate BY4741 wild type (score). image and cellpose_seg measure colony sizes from plate photographs, and viz draws plate heatmaps and histograms. The typical call sequence is read_gitter_dat(), read_echo_picklist(), merge_layout(), normalize_plate(), then score_plate().

Classes

NormalizationConfig	Knobs for the normalization stage.
ScoreReport	Top-level result of scoring one plate against its wild-type reference.
StrainScore	Per-strain aggregate: the analog of an SGAtools combined-file row.
CellposeSegConfig	Parameters for Cellpose-SAM plate segmentation.
PlateSegResult	Return bundle for a segmented plate: the gitter-style per-well table plus plate-level counts the classical path cannot report (raw instance count and off-grid contaminant colonies).

Functions

read_gitter_dat	Read a gitter .dat into a tidy frame: row, col, size, circularity, flags.
read_echo_picklist	Read an ECHO picklist into a layout frame: row, col, strain, volume_nl, well.
merge_layout	Attach strain + volume to each quantified colony by (row, col).
well_to_rowcol	Decode a plate well label ('B2', 'AA13') to 1-indexed (row, col).
normalize_plate	Run the full normalization on one plate's merged colony table.
score_plate	Score a normalized plate table (output of normalize_plate).
score_table	Flatten a ScoreReport into a per-strain DataFrame for CSV output.
quantify_plate_image	Quantify colonies on a standardized plate image into a gitter-style table.
quantify_plate_image_cellpose	Segment a backlit plate with Cellpose-SAM and quantify colonies onto the fitted array grid.
load_cellpose_model	Load the Cellpose-SAM generalist model (cpsam).
resolve_orientation	Return (merged, best_op, agreement).
volume_assay_metrics	One row per condition (volume) with the assay-quality metrics.
volume_position_confound	Detect whether the condition variable (volume) is confounded with plate position (row/col).
recommend_volume	Pick the best volume for a fitness screen.
zfactor	Z'-factor between two colony populations (assay separation window).