Led by Scott Kahn, PhD

Artificial Intelligence, Sleep Foundation Models, Recursive Sleep Intelligence, Multimodal Prediction, Digital Twins, and Adaptive Personalization

The NSMS National Computational Medicine & Digital Twin Center (aTwin™) is a multidisciplinary computational medicine and artificial intelligence center developing foundation models, Recursive Sleep Intelligence (RSI), multimodal AI, computational physiology, mechanistic simulation, and patient-specific digital twins for next-generation precision and autonomous healthcare.

Led by Scott Kahn, PhD, aTwin™ develops computational architectures capable of integrating electrophysiology, multimodal physiological sensing, molecular and laboratory data, medical imaging, environmental exposures, behavioral information, therapeutics, and longitudinal clinical trajectories into continuously evolving models of human health.

Within the NSMS precision sleep program, the Center leads development and integration of the Sleep Foundation Model (SFM), Recursive Sleep Intelligence (RSI), multimodal prediction, digital twin modeling, and adaptive personalization, creating the computational intelligence required to translate continuously measured physiology into patient-specific predictions and therapeutic strategies.

aTwin™ bridges:

• Artificial Intelligence

• Sleep Foundation Models (SFM)

• Recursive Sleep Intelligence (RSI)

• Digital Twin Systems

• Multimodal AI

• Computational Physiology

• Mechanistic Modeling

• Predictive Analytics

• Adaptive Personalization

• Causal & Mechanistic Reasoning

• Foundation Models

• Precision Therapeutics

• Autonomous Healthcare

Mission

To develop computational intelligence capable of continuously:

Integrating → Understanding → Modeling → Predicting → Personalizing → Adapting → Learning

across molecular, cellular, physiological, behavioral, environmental, and clinical scales.

The long-term objective is to transform healthcare from episodic assessment and population-level prediction toward continuous, patient-specific computational medicine.

Core Computational Programs

Sleep Foundation Model (SFM)

aTwin™ develops the Sleep Foundation Model (SFM) as a multimodal computational foundation for modeling restorative sleep, sleep physiology, health relationships, and longitudinal treatment response.

The SFM integrates information from:

• Polysomnography

• EEG and sleep microstructure

• Cardiovascular physiology

• PPG and HRV

• Respiratory physiology

• Oxygenation

• Circadian measurements

• Wearable sensors

• Movement and behavioral data

• Environmental exposures

• Clinical phenotypes

• Electronic health records

• Patient-reported outcomes

• Therapeutic-response data

Rather than relying solely on conventional sleep-stage labels, the SFM is designed to learn multidimensional representations of sleep physiology and its relationship to health.

Potential computational outputs include:

• Physiological state representations

• Sleep microstructure characterization

• Mechanistic phenotype identification

• Digital biomarker discovery

• Health-outcome prediction

• Treatment-response prediction

• Longitudinal trajectory modeling

• Patient similarity representations

• Personalized therapeutic targets

The SFM provides a common computational representation upon which downstream prediction, mechanistic reasoning, digital twins, and adaptive treatment systems can be developed.

Recursive Sleep Intelligence (RSI)

aTwin™ develops Recursive Sleep Intelligence (RSI) as a reasoning and adaptive-learning layer connecting physiological measurement, the SFM, patient-specific digital twins, and therapeutic response.

RSI integrates:

• Multimodal physiological inference

• Mechanistic reasoning

• Causal modeling

• Longitudinal learning

• Treatment-response analysis

• Uncertainty estimation

• Patient-specific adaptation

• Explainability

• Safety constraints

• Continuous model refinement

The RSI cycle follows:

Observe → Interpret → Reason → Predict → Select → Evaluate → Learn → Refine

As new physiological and therapeutic-response data become available, RSI updates its understanding of the individual while maintaining defined clinical and safety constraints.

Patient-Specific Digital Twins

aTwin™ develops dynamic computational representations of individual patients integrating multimodal information across time.

Digital twin inputs may include:

• Molecular biomarkers

• Laboratory measurements

• Medical imaging

• EEG and electrophysiology

• Cardiovascular physiology

• Respiratory physiology

• Sleep and circadian physiology

• Wearable sensing

• Environmental exposures

• Behavioral information

• Medication history

• Therapeutic interventions

• Electronic health records

• Longitudinal outcomes

Each digital twin is designed to continuously answer four fundamental questions:

What is happening now?
Why may it be happening?
What is likely to happen next?
How might alternative interventions change the trajectory?

The result is a continuously evolving computational representation of the individual rather than a static patient profile.

Multimodal AI & Foundation Models

aTwin™ develops AI systems capable of learning across heterogeneous biological and clinical information.

Research approaches include:

• Multimodal foundation models

• Transformer architectures

• Time-series learning

• Graph neural networks

• Physics-informed machine learning

• Causal inference

• Neurosymbolic AI

• Mechanistic-AI hybrid models

• Representation learning

• Continual learning

• Federated learning

• Uncertainty-aware inference

Multimodal models integrate:

EEG + Cardiac + Respiratory + Imaging + Biomarkers + Wearables + Environment + Behavior + EHR + Outcomes

to create richer representations of human physiological state than any single modality can provide independently.

Predictive Physiological Intelligence

aTwin™ develops computational models capable of estimating future physiological and clinical trajectories.

Applications include:

• Sleep-health outcome prediction

• Treatment-response prediction

• Clinical deterioration forecasting

• Cardiopulmonary instability

• Disease progression

• Recovery prediction

• Rehabilitation forecasting

• Surgical outcome prediction

• Chronic disease trajectories

• Physiological resilience

• Aging and longevity modeling

Predictions are coupled with uncertainty estimates and continuously updated as new patient information becomes available.

Adaptive Personalization

aTwin™ translates population-scale intelligence into continuously updated patient-specific models.

Adaptive personalization incorporates:

• Individual physiological baselines

• Mechanistic phenotypes

• Historical treatment response

• Current physiological state

• Patient-specific risk

• Predicted treatment response

• Treatment tolerability

• Behavioral context

• Environmental context

• Model confidence and uncertainty

The personalization cycle follows:

Measure → Model → Predict → Personalize → Intervene → Measure Response → Update

This creates the computational foundation for therapies that evolve according to measurable changes in the individual rather than remaining fixed over time.

Multiscale Computational Physiology

aTwin™ connects biological processes across:

Molecular → Cellular → Tissue → Organ → Physiological System → Whole Body → Environment

Research areas include:

• Neurophysiology

• Cardiovascular physiology

• Respiratory physiology

• Sleep physiology

• Circadian biology

• Autonomic regulation

• Metabolic physiology

• Immune and inflammatory dynamics

• Endocrine physiology

• Multiorgan interactions

• Physiological resilience

These models provide mechanistic context for AI-generated predictions and digital twin simulations.

Computational Pharmacology & Precision Therapeutics

aTwin™ develops patient-specific computational models for predicting therapeutic response.

Capabilities include:

• PK/PD modeling

• PBPK modeling

• Drug-response prediction

• Exposure-response modeling

• Patient-specific dosing simulation

• Treatment-response forecasting

• Benefit-risk modeling

• Combination-therapy simulation

• Virtual intervention comparison

• Longitudinal therapeutic optimization

Digital twins provide a computational environment in which alternative interventions can be modeled before translation into clinical action.

Self-Evolving Computational Intelligence

The integration of SFM + RSI + aTwin™ + continuous physiological feedback creates a self-evolving computational architecture.

The intelligence loop follows:

Multimodal Sensing → SFM Representation → RSI Reasoning → Digital Twin Simulation → Prediction → Adaptive Personalization → Therapeutic Decision → Physiological Response → Model Update

The system is designed to improve its patient-specific representation as longitudinal evidence accumulates while operating within predefined clinical, engineering, and safety constraints.

This architecture enables progression from:

Population Intelligence → Phenotype Intelligence → Patient Intelligence → Adaptive Personalization

Computational Technology Platforms

aTwin™: Patient-specific computational digital twin platform

SFM: Multimodal Sleep Foundation Model

RSI: Recursive Sleep Intelligence and adaptive reasoning architecture

aAI™: Multimodal artificial intelligence and foundation models

aPhysio™: Computational physiology and multiorgan modeling

aPredict™: Physiological and clinical trajectory prediction

aPersonalize™: Patient-specific adaptive personalization

aSim™: Mechanistic simulation and virtual intervention environment

aCausal™: Causal inference and mechanistic reasoning

aPKPD™: Computational pharmacology and therapeutic-response modeling

aCompute™: High-performance AI and distributed computational infrastructure

Integrated NSMS Computational Intelligence Ecosystem

aTwin™ serves as the AI, predictive intelligence, and patient-specific computational modeling layer connecting the broader NSMS autonomous healthcare ecosystem.

aSensors™ → aTwin™
Continuous multimodal physiological measurements

aArmor™ ↔ aTwin™
Wearable physiological states, predictions, personalization, and treatment-response feedback

aNeuro™ ↔ aTwin™
Neurophysiology, mechanistic biomarkers, neuromodulation modeling, and treatment response

aLab™ → aTwin™
Molecular, laboratory, biomarker, and mechanistic data

aEyes™ → aTwin™
Multiscale imaging and quantitative diagnostic information

aData™ ↔ aTwin™
Standardized clinical data, interoperability, evidence, provenance, and longitudinal outcomes

aVerify™ ↔ aTwin™
AI/model requirements, interfaces, traceability, verification, integration, and V&V

aValidate™ → aTwin™
Clinical outcomes, external validation, and prospective evidence

DrRobots™ ↔ aTwin™
Digital-to-physical integration supporting clinician-supervised autonomous healthcare systems

aLaunch™ ← aTwin™
Translation of validated AI and digital twin technologies toward partnerships and healthcare deployment

Together, these capabilities establish the computational pathway:

Sense → Integrate → Model → Reason → Simulate → Predict → Personalize → Validate → Learn

R&D / Computational Domains

• Artificial Intelligence

• Sleep Foundation Models

• Recursive Sleep Intelligence

• Digital Twin Systems

• Multimodal AI

• Computational Medicine

• Computational Physiology

• Predictive Analytics

• Foundation Models

• Adaptive Personalization

• Mechanistic Modeling

• Causal Inference

• Neurosymbolic AI

• Reinforcement Learning

• Computational Pharmacology

• Systems Biology

• Computational Neuroscience

• Precision Therapeutics

• Self-Evolving Intelligence

• Autonomous Healthcare

Leadership

Scott Kahn, PhD
Director, NSMS National Computational Medicine & Digital Twin Center (aTwin™)
AI, SFM, RSI & Digital Twin Lead

Scott Kahn, PhD, leads NSMS artificial intelligence, Sleep Foundation Model (SFM), Recursive Sleep Intelligence (RSI), multimodal AI, predictive modeling, adaptive personalization, and digital twin activities.

His work focuses on integrating mechanistic biology, computational physiology, multimodal artificial intelligence, longitudinal patient data, and continuous physiological measurements into dynamic models capable of representing individual health states and predicting how those states may evolve.

Under his leadership, aTwin™ provides the computational intelligence required to connect multimodal sensing → SFM → RSI → patient-specific digital twins → prediction → adaptive personalization → treatment-response learning.

The Future of Computational Medicine

The future of healthcare will require more than AI systems that classify disease or predict population-level risk.

It will require computational intelligence capable of continuously understanding the individual—integrating biological mechanisms, multimodal physiology, longitudinal history, environmental context, and therapeutic response into an evolving model of human health.

aTwin™ is developing that foundation through the integration of SFM, RSI, multimodal AI, predictive modeling, and patient-specific digital twins.

Understand the Physiology. Model the Individual. Predict the Trajectory. Personalize the Intervention. Learn from the Response.

aTwin™ — Computational Intelligence for Precision and Autonomous Medicine.