🧬 Endo-ML: Machine Learning for Endometriosis Biomarker Discovery

A multimodal bioinformatics and machine learning project for identifying biomarkers and predicting endometriosis status using RNA-Seq and DNA methylation data.

🚀 Overview & Motivation

Endometriosis is an under-researched gynecological disorder affecting an estimated 5–10 % of women.
Current diagnostics are invasive and often delayed.

This project applies bioinformatics and ML techniques to support non-invasive diagnosis and molecular understanding of endometriosis.

This study combines:

• Differential gene expression analysis (RNA-Seq)
• Differential methylation analysis (e.g., MBD-Seq)
• Multi-omics integration
• Supervised ML models for disease prediction (Two machine learning architectures)
• Feature selection and biological interpretation (GO / pathway enrichment)

This project demonstrates omics data processing, differential analysis, functional enrichment, feature engineering, model development (ensemble ML modeling), and biological reasoning.

🧠 Workflow Summary

1. RNA-Seq analysis (DESeq2)

   • Filtering low-count genes, normalization (median of ratios), log2 fold-change shrinkage (apeglm)
   • 100 significant DEGs (padj < 0.05): 83 downregulated, 17 upregulated
   • Top genes: COL1A1, TENM2, IGF2, KLF2P1

2. Methylation analysis (edgeR)

   • Normalization via TMM, GLM + LRT testing
   • 27 significant DMRs: 11 hypo-, 16 hyper-methylated
   • Top regions: GGA1, TOM1L1, RPL5P10, NT5DC1P1

3. Functional Enrichment (R / Bioconductor)

   • Tools: clusterProfiler, ReactomePA, GOstats, g:Profiler
   • Significant Reactome pathways: collagen formation, IGF signaling
   • Key gene hits: IGFBP3, COL1A1, GGA1

4. Machine Learning (Python / scikit-learn)

   • Model 1: Two logistic regressions (RNA + Methylation) → Meta ensemble
     • Combined probability averaging
     • Best performance: Accuracy = 0.92, AUC = 0.94
     • Key predictive genes: KLF2P1, SCAF1, IGF2, CDCA2
   • Model 2: Stacked classifier (Logistic + Decision Tree + SVM)
     • Combined DEGs + DMRs
     • Accuracy = 0.83, AUC = 0.83

🧰 Tech Stack

📁 Repository Content

Diff_Analysis_mbdseq.R → MBD-seq differential methylation analysis
GoTermEnrichment.Rmd → GO enrichment pipeline
GoTermEnrichment.html → Rendered enrichment report
ML_model1.ipynb / .html → Machine learning model 1
ML_model2.ipynb / .html → Machine learning model 2
MethylationAnnotation.R → Methylation site annotation
rnaseq.Rmd / .html → RNA-seq analysis pipeline
Report_Group3.pdf → Final report
ReadMe.docx.pdf → Documentation
README.md → Project overview (this file)

📈 Model Performance & Results

The objective of this project was to build a high-performance machine learning classifier capable of predicting endometriosis status using integrated multi-omics data (RNA-Seq + Methylation).
The results clearly show that ensemble learning and meta-model integration significantly outperform baseline models trained on individual data types.

🔹 1. Ensemble Learning vs. Baseline Models

A Stacked Classifier was developed by combining Logistic Regression, Decision Tree, and SVM base learners.
It achieved the highest and most balanced performance across all metrics.

Conclusion:
The ensemble approach boosts predictive performance and improves stability, increasing the F1-score and AUC to 0.83.

🔹 2. Multi-Omics Meta-Model Integration

To further enhance prediction, a Meta-Model was trained by combining features derived from both RNA expression and DNA methylation datasets.
This model substantially outperformed any single-omics model.

Conclusion:

Integrating multi-omics features delivers state-of-the-art performance, achieving an F1-score of 0.91 and AUC of 0.94, demonstrating the effectiveness of data fusion in endometriosis biomarker prediction.

📊 Overall Insight:

Multi-omics integration and ensemble learning dramatically improve classification accuracy and robustness, validating this approach for complex biological datasets.

Highlighted features:

• IGF2 (RNA downregulated, literature-supported biomarker)
• CDCA2 (hypermethylated region)
• IGFBP3 (methylation linked to lesion development)
• KLF2P1, SCAF1 (novel candidates)

🔬 Limitations & Future Work

• Small overlapping sample size between RNA and MBD datasets
• Limited external validation due to data availability
• Future directions:
  • Larger cohort validation
  • Bootstrapping for variance reduction
  • Deep learning integration (autoencoders / multimodal fusion)
  • Expanded omics integration (proteomics, metabolomics)
  • Web API or Streamlit interface for inference

⚙️ Usage

# Clone repository
git clone https://github.com/astriksar/endo_ml.git
cd endo_ml

📄 License & Citation

This project is released under the MIT License — see the LICENSE file for details.
You are welcome to use, modify, and build upon this code for academic or research purposes.

If you use this repository or its results in your own research, please cite as follows:

Sarukhanyan, A., Forsythe, C., & Lapucha, J. (2025).
Endo-ML: Integrated Machine Learning and Multi-Omics Analysis for Endometriosis Biomarker Discovery.
GitHub Repository: https://github.com/astriksar/endo_ml

Alternatively, include this citation in BibTeX format:

@misc{SarForLap2025EndoML,
  author       = {Astghik Sarukhanyan and Caroline Forsythe and Julia Lapucha},
  title        = {Endo-ML: Integrated Machine Learning and Multi-Omics Analysis for Endometriosis Biomarker Discovery},
  year         = {2025},
  publisher    = {GitHub},
  journal      = {GitHub repository},
  howpublished = {\url{https://github.com/astriksar/endo_ml}},
  license      = {MIT}
}
---