AML Venetoclax Resistance

Code and resources to support the main bootstrap-based L1-logistic-regression analysis used in this study.

---

Overview

This repository contains scripts and example data for the bootstrap-based feature selection analysis described in the paper.
Its purpose is to improve transparency and reproducibility of the main computational results.

---

Repository Contents

run_bootstrap_iteration.py

Runs a single bootstrap iteration of L1-regularized logistic regression.

This step:

• resamples the input data with replacement,
• adds small Gaussian noise,
• fits an L1-penalized logistic regression model,
• returns the selected-feature mask (|coef| > epsilon), coefficients, and out-of-bag (OOB) sample indices.

fit_lasso_logistic_bootstrap.py

Runs repeated bootstrap iterations in parallel.

This step:

• performs n_bootstrap iterations,
• collects coefficients across iterations,
• returns a coefficient matrix of shape (n_bootstrap × n_features) and the OOB index list.

summarize_bootstrap_coefficients.py

Summarizes bootstrap results across iterations.

This step:

• computes feature-selection frequency,
• calculates coefficient means and standard deviations,
• identifies features passing a user-defined selection threshold,
• returns summary statistics for downstream interpretation.

reproduce_main_analysis.py

Runs the main synthetic-data example from start to finish.

This script:

• loads the example input files,
• runs bootstrap-based L1-logistic regression,
• summarizes feature stability across bootstrap iterations,
• writes analysis outputs to the results/ directory.

---

Analysis Outline

The core workflow is:

1. Prepare a processed feature matrix X and binary label vector y.
2. Run bootstrap-based L1-logistic regression across repeated resamples.
3. Summarize selection frequency and coefficient stability across bootstrap iterations.
4. Identify robustly selected features.

---

Installation

Clone the repository and create a clean virtual environment:

git clone https://github.com/kenflab/aml-venetoclax-resistance
cd aml-venetoclax-resistance

python3 -m venv .venv
source .venv/bin/activate   # macOS / Linux
# .venv\Scripts\activate    # Windows

pip install --upgrade pip
pip install -r requirements.txt

Run the Main Reproducible Example

To run the main synthetic-data analysis:

python reproduce_main_analysis.py

This command writes the following files to the results/ directory:

• bootstrap_coefficients.csv
• bootstrap_feature_summary.csv
• selected_features.json
• analysis_metadata.json

---

Example Usage

gene_names = X.columns
n_bootstrap = 10000
epsilon = 0.01
random_state = 2025
n_jobs = -1
threshold_ratio = 0.8

params = {
    "C": 10,
    "class_weight": "balanced",
    "max_iter": 10000,
}

coef_matrix, oob_indices = fit_lasso_logistic_bootstrap(
    X.values,
    y,
    gene_names=gene_names,
    n_bootstrap=n_bootstrap,
    epsilon=epsilon,
    n_jobs=n_jobs,
    random_state=random_state,
    params=params,
)

summary = summarize_bootstrap_coefficients(
    coef_matrix=coef_matrix,
    gene_names=gene_names,
    threshold_ratio=threshold_ratio,
)

Input Data Assumptions

• X: A numeric matrix where

  • Rows represent samples
  • Columns represent features (genes)
  • Values are Transcripts Per Million (TPM), then log-transformed, and Z-score standardized per gene across samples.

• y: A binary list or array of labels (0 and 1), corresponding to the sample classes.

---

Example Data

This repository includes synthetic example data for demonstration and code verification only.

• The example data were generated using NumPy random numbers with a fixed seed (42).

• No real patient or biological data are included in this repository.

• Files are located in the data/ folder:

  • data/X_sample.csv — Processed feature matrix (rows = samples, columns = genes).
    Values are TPM-normalized, then log-transformed, and Z-score standardized per gene across samples.
  • data/y_sample.csv — Labels (sample_id, label) with binary classes 0/1.

Shapes

• X_sample.csv: 30 × 50
• y_sample.csv: 30 × 2 (sample_id, label)

---

Quick Load Example

import pandas as pd

# Load X (processed) with sample IDs as index
X = pd.read_csv("data/X_sample.csv", index_col=0)

# Load y and align to X.index
y_df = pd.read_csv("data/y_sample.csv")  # columns: sample_id, label
y = y_df.set_index("sample_id").loc[X.index, "label"].values

print(X.shape, y.shape)
print(X.head(3))
print(y[:10])

---

Notes

This repository is intended to support the analysis presented in the manuscript. It is not packaged as a general-purpose software tool, but rather as a focused, minimal resource for reproducing the main bootstrap-based feature selection workflow.

For the original study data, please refer to the manuscript and its data availability statement.