Model Training and Evaluation Pipeline

This repository contains the code used to train, evaluate, and post-process machine learning models across multiple configurations and random seeds.

The pipeline is designed to support large-scale experimentation and comparative analysis across disease cohorts, timepoints, and model settings.

The final trained models and the complete MISPA dataset are available for download at the following link: https://drive.google.com/drive/u/1/folders/1jbK-60344zbNYwgxpiV_oz9djLEKwKF6

---

Overview

• Trains models across multiple configurations
• Uses 100 random seeds per configuration for robustness
• Produces final trained models and aggregated evaluation results
• Supports downstream analysis grouped by disease cohort, train/test timepoint, and model configuration

---

Running the Pipeline

To train models for all configurations over 100 random seeds, run:

python main.py

This will generate all intermediate outputs and trained models required for analysis.

Requirements All required Python packages are listed in requirements.txt. Install dependencies with:

pip install -r requirements.txt

These models correspond to the full set of configurations and random seeds used in the study.

Post-processing and Analysis The notebook final_leidos.ipynb is used to post-process and analyze model outputs. Results are aggregated and grouped by:

Disease cohort, Train/test timepoint, Model configuration. This notebook generates the final summaries used for evaluation.

Reproducibility All experiments are run using fixed random seeds Each configuration is evaluated over 100 independent seeds Trained models and outputs are saved for reproducibility and downstream analysis Notes Ensure sufficient compute and storage before running the full pipeline Downloading pre-trained models is recommended if retraining is not feasible