Related Experiment Video
Updated: May 10, 2026
DeepOmicsAE: Representing Signaling Modules in Alzheimer's Disease with Deep Learning Analysis of Proteomics, Metabolomics, and Clinical Data
Published on: December 15, 2023
A modular deep learning architecture for interpretable disease prediction across tabular clinical and biometric
Vijay U Rathod1, Siddhesh Sanjay Amrutkar1, Kirti A Patil2
1Department of CSE (Artificial Intelligence and Machine Learning), Vishwakarma Institute of Technology (Affiliated to Savitribai Phule Pune University, Pune), Pune, Maharashtra, India.
This study introduces a unified deep learning framework for accurate disease prediction across diverse clinical datasets, enhancing early diagnosis and clinical decision support. The adaptable model generalizes well, offering improved reliability for real-world healthcare applications.
Area of Science:
- Biomedical Informatics
- Machine Learning
- Computational Biology
Background:
- Deep learning models for disease prediction are often disease-specific and struggle with generalization across varied biomedical datasets.
- Existing approaches can be computationally intensive, limiting their practical application in clinical decision support.
- There is a need for a unified, adaptable framework for accurate and interpretable disease prediction.
Purpose of the Study:
- To propose a unified, dataset-aware deep learning framework for accurate and interpretable disease prediction.
- To enable generalization across heterogeneous clinical datasets by adapting model selection based on dataset characteristics.
- To improve the reliability and practical applicability of deep learning in multi-disease prediction.
Main Methods:
- Developed a modular deep learning framework integrating MLP, 1D CNN, FT-Transformer, autoencoders, and ensemble strategies.
- Implemented dataset-aware model selection based on feature dimensionality, sample size, and class imbalance.
- Employed robust preprocessing, fold-safe feature selection, and nested cross-validation for reliable performance evaluation.
Main Results:
- The framework demonstrated competitive predictive performance across three diverse datasets (heart disease, diabetes, Parkinson's disease).
- An FT-Transformer + autoencoder ensemble achieved an AUC of 0.8980 for heart disease prediction.
- A CNN + Autoencoder ensemble achieved an AUC of 0.8451 for diabetes classification, with MLP achieving perfect specificity for Parkinson's disease detection.
Conclusions:
- The proposed framework offers a scalable and interpretable solution for multi-disease prediction.
- The dataset-aware approach enhances model generalization and reliability in real-world healthcare settings.
- This work contributes to improving early diagnosis and clinical decision-support systems through advanced deep learning techniques.