nuclear-table-tools

Focused dataframe operations for object-level microscopy measurements. The package covers filtering, feature selection, correlation pruning, joins between classical and learned representations, and stateful pipeline composition.

Install

python -m pip install .

Example

from nuclear_table_tools import drop_high_corr_columns, drop_low_unique_columns

feature_columns = [column for column in table if column.startswith("DAPI_")]
table = drop_low_unique_columns(table, feature_columns, unique_threshold=10)
feature_columns = [column for column in feature_columns if column in table]
table = drop_high_corr_columns(table, feature_columns, corr_threshold=0.95)

Join learned representations without losing the object-level identity keys:

from nuclear_table_tools import merge_with_VAE_df

combined = merge_with_VAE_df(table, latent_table)

Compose project-specific derivations and row filters while retaining a readable dataframe workflow:

from nuclear_table_tools import AdvancedOperator

operator = AdvancedOperator(combined)
operator.add_column("DAPI_mean_to_area", combined["DAPI_mean"] / combined["area"])
operator.filter_rows(lambda frame: frame["DAPI_mean_to_area"].notna())
analysis_table = operator.get_dataframe()

Transformations preserve row identities and keep metadata columns alongside the selected numeric feature matrix.