• SOUMYA RANJAN MISHRA
• SACHIKANTA DASH
• GVS NARAYANA

Diabetes remains a critical global health issue, necessitating advanced methodologies to unravel its complex etiology and enhance predictive capabilities for better management and therapeutic interventions. Existing approaches to diabetes analysis predominantly rely on centralized models, which are limited by privacy concerns, data heterogeneity, and the inability to capture comprehensive causal relationships among multi-omics data samples. To address these limitations, we introduce a novel suite of models leveraging federated learning and causal inference techniques: the FedOmics Causal Network (FOCN), Federated Multi-Omics Variational Autoencoder (FMO-VAE), and Causal Omics Pathway Inference (COPI) Process. FOCN utilizes federated learning to train a deep causal network on distributed multi-omics datasets from various medical institutions. By integrating genetic, proteomic, and metabolomic data, FOCN infers causal relationships between molecular features and diabetes outcomes, enhancing our understanding of disease mechanisms and potential therapeutic targets. The model's federated architecture ensures data privacy while achieving a notable 10% increase in the Area under the Curve (AUC) for diabetes risk prediction compared to baseline models. The FMO-VAE model employs a variational autoencoder trained via federated learning, capturing latent representations of multi-omics data while preserving privacy across institutions for different scenarios. This approach harmonizes data from diverse sources without sharing raw data, significantly reducing reconstruction error by 20% compared to centralized variational autoencoder models. The latent space representation learned by FMO-VAE facilitates knowledge transfer and enhances the integrative analysis of multi-omics data samples. COPI integrates multi-omics data with causal inference models to infer causal pathways involved in diabetes progression. Utilizing Bayesian Structural Causal Models (BSCMs), COPI identifies causal relationships between molecular pathways and clinical outcomes, revealing novel insights into disease etiology. For instance, COPI uncovers a causal link between dysregulated lipid metabolism pathways and insulin resistance, highlighting potential therapeutic targets for intervention operations. This work demonstrates significant advancements in diabetes research by addressing critical limitations of existing methodologies. The proposed models not only improve predictive accuracy and data privacy but also provide mechanistic insights into diabetes pathogenesis. By elucidating causal relationships and integrating diverse multi-omics data, these models offer a robust framework for future research and clinical applications in diabetes management and treatment.

Diabetes, Multi-Omics, Federated Learning, Causal Inference, Variational Autoencoder

"Iterative Model Design for Diabetes Analysis Using FedOmics Causal Network and Federated Multi-Omics Variational Autoencoder", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.11, Issue 6, page no.g295-g301, June-2024, Available :http://www.jetir.org/papers/JETIR2406635.pdf

"Iterative Model Design for Diabetes Analysis Using FedOmics Causal Network and Federated Multi-Omics Variational Autoencoder", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.11, Issue 6, page no. ppg295-g301, June-2024, Available at : http://www.jetir.org/papers/JETIR2406635.pdf