lno-prescreener 0.1.3

A multiple hypothesis testing tool for pre-screening high-dimensional data using the Leave-n-Out approach.

lno-prescreener

A high-performance, multiple hypothesis testing tool for pre-screening factor selection in high-dimensional data using the Leave-n-Out (LnO) approach.

Designed for researchers working with metabolomic, genomic, or clinical datasets, this library provides a robust, multi-core pipeline to filter out false discoveries using adaptive univariate testing and combinatorial resampling.

FEATURES

• Scientifically Rigorous: Implements an adaptive decision tree (Anderson-Darling, Levene, KS, Mann-Whitney, t-tests) perfectly aligned with established LnO methodology.
• Highly Flexible: Works with any Pandas DataFrame. You can define your own target columns, cohort labels, and index parameters.
• Multi-Core Processing: Utilizes joblib parallel processing to evaluate thousands of features simultaneously, cutting compute times from hours to seconds.
• Missing Data Handling: Automatically purges incomplete features to maintain strict data integrity.
• Defensive Design: Explicit input validation prevents obscure math crashes and provides clear, human-readable errors if your data is misconfigured.

THE DETAILS OF THE LNO PRESCREENING METHOD CAN BE FOUND IN THE PAPER: Halil Arici, Fatir A. Qureshi, Jay Mulmule, Juergen Hahn,

Multiple hypothesis testing for pre-screening of factor selection for classification of high-dimensional data,

Journal of Process Control, Volume 152, 2025, 103469, ISSN 0959-1524, https://doi.org/10.1016/j.jprocont.2025.103469.

(https://www.sciencedirect.com/science/article/pii/S0959152425000976)

PLEASE CITE THE ORIGINAL PAPER IF THIS PACKAGE IS USED FOR YOUR RESEARCH OR PAPER.

INSTALLATION

You can install lno-prescreener directly from PyPI: pip install lno-prescreener

QUICK START

Here is a concise example of how to load your data and run the pre-screener pipeline.

from lno_prescreener import LnOPrescreener

Initialize the pre-screener

screener = LnOPrescreener( data_input="sqrt_transformed_data.csv", target_variable="Diagnosis", case_label="ASD", control_label="TD", index_col=0 #(int or None) The column number to use as the row labels (default: 0). Set to None if your data does not have an ID column. )

Run the multiple hypothesis testing pipeline

results = screener.multiple_hypothesis_testing( run_l2o=False, run_l3o=True, fdr_threshold=0.1, p_val_threshold=0.05 )

Save the full comprehensive results to CSV

results.to_csv("prescreening_results.csv", index=False)

API REFERENCE

Initialization LnOPrescreener(data_input, target_variable='Diagnosis', case_label='ASD', control_label='TD', index_col=0)

• data_input: (str or DataFrame) Path to your CSV file, or a pre-loaded Pandas DataFrame.
• target_variable: (str) The exact name of the column containing your class labels.
• case_label: (str) The exact string used to identify cases in the target column.
• control_label: (str) The exact string used to identify controls in the target column.
• index_col: (int or None) The column number to use as the row labels (default: 0). Set to None if your data does not have an ID column.

Execution screener.multiple_hypothesis_testing(run_l2o=False, run_l3o=False, p_val_threshold=0.05, fdr_threshold=0.10)

• run_l2o: (bool) If True, runs the Leave-2-Out permutation tests. Note: Computationally expensive for large N.
• run_l3o: (bool) If True, runs the Leave-3-Out permutation tests. Note: Highly computationally expensive for large N.
• p_val_threshold: (float) The alpha threshold for the baseline and adaptive univariate tests (default: 0.05).
• fdr_threshold: (float) The maximum allowable False Discovery Rate for a feature to be flagged as significant (default: 0.10).

OUTPUT DATAFRAME The multiple_hypothesis_testing method returns a rich Pandas DataFrame containing:

• The original and shifted p-values.
• The specific statistical test applied by the adaptive decision tree.
• Bonferroni and Benjamini-Hochberg corrected p-values.
• Raw FDR scores and boolean Significant flags for Baseline, L1O, L2O, and L3O.