WIMOAD: Stacking Ensemble and Weighted Multi-Omics Integration for Alzheimer's Disease Diagnosis

WIMOAD a stacking ensemble and weighted integration of multi-omics data for AD diagnosis. WIMOAD synergistically leverages specialized classifiers for patients' paired gene expression and methylation data for multi-stage classification. The resulting scores of classifiers were then stacked for meta-learning performance improvement. The prediction results of two distinct meta-models were integrated with optimized weights for the final decision-making of the model, providing higher performance than using single omics only.

Table of Contents (Under construction)

• Overview
• Installation
• Data Format
• Usage
• Using WIMOAD on Your Own Data
• Contact
• Publication

Overview

WIMOAD is organized around two stages, each usable independently:

• Meta-learning (stacking) — runner.py, main.py, parallel.py/parallel_loo.py, model_config.py, data_loader.py, metrics.py: for each omics branch, trains base classifiers, stacks their out-of-fold predictions with several meta-models under nested cross-validation, and reports evaluation metrics.
• Integration (weighted fusion) — integration.py: given each branch's best stacking model, searches the expression/methylation weight that maximizes the fused prediction's AUC.

This repository ships the modeling framework only; it does not include or require any specific dataset.

Installation

1. Clone the WIMOAD git repository:

git clone https://github.com/wan-mlab/WIMOAD.git
cd WIMOAD

2. Create a new conda environment:

conda create -n wimoad python=3.9
conda activate wimoad

3. Install Python dependencies:

pip install -r requirements.txt

Data Format

WIMOAD expects one CSV per omics branch, each with:

• RID: a unique sample identifier (used as the index)
• DX_bl: the raw diagnosis label for that sample
• remaining columns: feature values for that branch (gene expression or methylation levels)

Raw DX_bl values are remapped to a binary label per task using the label_map defined for that task in configs/tasks.yaml; any samples with a value not covered by the mapping's binary classes are dropped. No other assumptions are made about the data — feature columns, sample counts, and diagnosis coding are all task/dataset-specific and configured in configs/tasks.yaml.

Usage

Run the default ca task for both omics branches with leave-one-out CV:

python runner.py

Run a specific branch, CV strategy, and output directory:

python runner.py --group ca --omics expression --cv KFold --output-dir results

Core files:

• configs/tasks.yaml: task groups, label maps, and base-model hyperparameters
• data_loader.py: CSV loading, label remapping, and feature selection
• model_config.py: meta models and sklearn estimator factories built from configs/tasks.yaml
• parallel.py / parallel_loo.py: nested-CV stacking training (KFold and leave-one-out outer loops)
• metrics.py: classification metrics computed per stacking run
• runner.py: command-line stacking workflow
• integration.py: weighted multi-omics fusion — best-model selection per branch and fusion weight search

Using WIMOAD on Your Own Data

1. Prepare one CSV per omics branch in the format described in Data Format.
2. Add a task entry to configs/tasks.yaml: point expression_file/methylation_file at your CSVs, define the label_map for your diagnosis coding, and set base-classifier hyperparameters for each branch.
3. Run python runner.py --group <your_task> to train and evaluate the stacking ensembles for that task.
4. Use integration.py to select each branch's best model and search the fusion weight over the resulting predictions.

Contact

If you have any questions, comments, or would like to report a bug, please contact haxiao@unmc.edu

Publication

Xiao, H.; Wang, J.; Wan, S. WIMOAD: Weighted Integration of Multi-Omics data for Alzheimer's Disease (AD) Diagnosis. bioRxiv 2024.09.25.614862, https://www.biorxiv.org/content/10.1101/2024.09.25.614862v1