This paper presents a simulation-based, AI-driven personal health monitoring system that integrates multi-omic and physiological data to enable early anomaly detection and emergency response. The system features a high-fidelity synthetic patient generator which models 1,000 profiles across 60 days, incorporating disease pathways (e.g., sepsis, diabetes and cardiac) based on KEGG annotations and variables such as age, gender, genetic risk, circadian rhythm, and environmental stressors. Later, this simulated physiological and omic data feed into a multiple-stage diagnostic pipeline: an unsupervised Isolation Forest model (92.3%), followed by an unsupervised LSTM Autoencoder (90.1%), and finally a supervised LSTM model (93.8%) for multi-class health classification. The system includes a data security module utilizing AES-256 encryption, a client-server architecture linking a Unity simulation frontend with a Python backend, and deployment on Microsoft Azure. Emergency alerts are sent through Twilio SMS with GPS integration to identify and contact the nearest emergency center to the current location, while real-time clinical updates are delivered to a Firebase-powered mobile application designed for healthcare providers. All experiments and metrics reported in this system derive exclusively from simulation with synthetic patients; thus, no real/physical devices or clinical data were used. This end-to-end platform offers a novel, scalable proof-of-concept that unifies biosimulation, wearable biomedical systems, multi-omic AI pipelines, and real-time emergency alerting within a secure, cloud-integrated architecture.
Celik et al. (Thu,) studied this question.