Genomic Deep Learning: A Systematic Review on Reproducible Ensemble Deep Learning Framework for Robust Cardiac and Pulmonary Disease Classification

Genomic deep learning has become an increasingly important approach for disease risk prediction using high-dimensional gene expression and SNP derived features. However, existing deep learning studies in genomics often suffer from limited reproducibility, sensitivity to dataset imbalance and instability across cohorts. This work proposes a reproducible ensemble deep learning framework for robust classification of cardiac and pulmonary diseases, leveraging standardized preprocessing, biologically grounded feature encoding and ensemble modeling. The framework is validated using publicly available genomic datasets including GSE90076 and GSE31276, demonstrating improved robustness, generalization and interpretability compared to single model approaches. Objectives: Primary: Systematically design and evaluate a reproducible ensemble deep learning framework for genomic based classification of cardiac and pulmonary diseases. Secondary: 1. Compare ensemble model performance against individual deep learning and traditional machine learning baselines 2. Analyze the impact of ensemble strategies and feature encoding on classification robustness 3. Assess cross-dataset generalization between cardiac and pulmonary disease cohorts 4. Evaluate methodological transparency and reproducibility of the proposed framework 5. Identify biologically meaningful gene features contributing to disease prediction Expected Outcomes: Deliverables: 1. A fully reproducible genomic deep learning pipeline for cardiac and pulmonary disease classification 2. Quantitative performance benchmarking of ensemble models versus single model baselines 3. Identification of stable, disease-relevant genomic features 4. Cross cohort validation results demonstrating robustness and scalability 5. Reusable framework supporting future genomic deep learning research Impact: 1. Evidence-based guidance for applying ensemble deep learning in genomic disease prediction 2. Improved robustness and reproducibility of deep learning models in cardiac and pulmonary genomics 3. Benchmark methodology for future SNP and gene-expression based classification studies 4. Support for translational research through interpretable and generalizable genomic predictors