coSMicQC in a nutshell

This notebook demonstrates various capabilities of coSMicQC using examples.

import pathlib

import pandas as pd
from cytodataframe import CytoDataFrame

import cosmicqc

from importlib.metadata import version

version("cytodataframe")

'0.3.0'

# set a path for the parquet-based dataset
# (in this case, CellProfiler SQLite data processed by CytoTable)
data_path = (
    "../../../tests/data/cytotable/NF1_cellpainting_data/"
    "Plate_2_with_image_data.parquet"
)

# set a context directory for images associated with the dataset
image_context_dir = pathlib.Path(data_path).parent / "Plate_2_images"
mask_context_dir = pathlib.Path(data_path).parent / "Plate_2_masks"

# create a cosmicqc CytoDataFrame (single-cell DataFrame)
scdf = CytoDataFrame(
    data=data_path,
    data_context_dir=image_context_dir,
    data_mask_context_dir=mask_context_dir,
)

# display the dataframe
scdf

Static snapshot (for non-interactive view)

	Metadata_ImageNumber	Image_Metadata_Plate_x	Metadata_number_of_singlecells	Image_Metadata_Site_x	Image_Metadata_Well_x	Metadata_Cells_Number_Object_Number	Metadata_Cytoplasm_Parent_Cells	Metadata_Cytoplasm_Parent_Nuclei	Metadata_Nuclei_Number_Object_Number	Cytoplasm_AreaShape_Area	...	Image_Threshold_SumOfEntropies_Cells	Image_Threshold_SumOfEntropies_Nuclei	Image_Threshold_WeightedVariance_Cells	Image_Threshold_WeightedVariance_Nuclei	Image_URL_DAPI	Image_URL_GFP	Image_URL_RFP	Image_Width_DAPI	Image_Width_GFP	Image_Width_RFP
0	1	Plate_2	44	1	A12	1	1	2	2	21024.0	...	-12.181288	-11.699993	0.992624	0.657791				1224	1224	1224
1	1	Plate_2	44	1	A12	4	4	7	7	12754.0	...	-12.181288	-11.699993	0.992624	0.657791				1224	1224	1224
2	1	Plate_2	44	1	A12	7	7	10	10	23976.0	...	-12.181288	-11.699993	0.992624	0.657791				1224	1224	1224
3	1	Plate_2	44	1	A12	8	8	12	12	19374.0	...	-12.181288	-11.699993	0.992624	0.657791				1224	1224	1224
4	1	Plate_2	44	1	A12	9	9	13	13	27385.0	...	-12.181288	-11.699993	0.992624	0.657791				1224	1224	1224
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
1709	128	Plate_2	59	4	H7	10	10	14	14	24942.0	...	-12.566582	-11.633043	1.624310	0.545186				1224	1224	1224
1710	128	Plate_2	59	4	H7	11	11	15	15	6627.0	...	-12.566582	-11.633043	1.624310	0.545186				1224	1224	1224
1711	128	Plate_2	59	4	H7	12	12	16	16	11216.0	...	-12.566582	-11.633043	1.624310	0.545186				1224	1224	1224
1712	128	Plate_2	59	4	H7	13	13	17	17	15279.0	...	-12.566582	-11.633043	1.624310	0.545186				1224	1224	1224
1713	128	Plate_2	59	4	H7	14	14	20	20	7106.0	...	-12.566582	-11.633043	1.624310	0.545186				1224	1224	1224

1714 rows × 2076 columns

bbox_cols = [
    col for col in scdf.columns if "bbox" in col.lower() or "box" in col.lower()
]

print("bbox_col:", bbox_cols)
print("bbox_cols:")
for col in bbox_cols:
    print(col)

# Identify which rows include outliers for a given threshold definition
# which references a column name and a z-score number which is considered
# the limit.
cosmicqc.analyze.identify_outliers(
    df=scdf,
    feature_thresholds={"Nuclei_AreaShape_Area": -1},
).sort_values()

# Show the number of outliers given a column name and a specified threshold
# via the `feature_thresholds` parameter and the `find_outliers` function.
cosmicqc.analyze.find_outliers(
    df=scdf,
    metadata_columns=["Metadata_ImageNumber", "Image_Metadata_Plate_x"],
    feature_thresholds={"Nuclei_AreaShape_Area": -1},
)

# create a labeled dataset which includes z-scores and whether those scores
# are interpreted as outliers or inliers. We use pre-defined threshold sets
# loaded from defaults (cosmicqc can accept user-defined thresholds too!).
labeled_scdf = cosmicqc.analyze.label_outliers(
    df=scdf, include_threshold_scores=True, feature_thresholds="large_nuclei"
)
labeled_scdf

# show cropped images through CytoDataFrame from the dataset to help analyze outliers
# labeled_scdf._enbable_debug_mode()
labeled_scdf.sort_values(by="Metadata_cqc_large_nuclei_is_outlier", ascending=False)[
    [
        "Metadata_ImageNumber",
        "Metadata_Cells_Number_Object_Number",
        "Metadata_cqc_large_nuclei_is_outlier",
        "Image_FileName_GFP",
        "Image_FileName_RFP",
        "Image_FileName_DAPI",
    ]
]

# One can convert from cosmicqc.CytoDataFrame to pd.DataFrame's
# (when or if needed!)
df = pd.DataFrame(scdf)
print(type(df))
df