Digital Twin Clinical Agent

AI-powered patient digital twin creation for clinical trial simulation, treatment outcome prediction, and personalized medicine using real-world data and multi-omics integration.

FridrichMethod 300715d 7.3 KB Updated Sep 17, 2026

Install

Press C to copy

$ npx skillmds@latest add fridrichmethod/digital-twin-clinical-agent

Installs into ~/.claude/skills (or .claude/skills inside a project). Restart Claude Code to pick it up.

Download SKILL.md

＋ Add to collection

README badge

[![SkillMD](https://skillmd.com/badge/fridrichmethod/digital-twin-clinical-agent.svg)](https://skillmd.com/skills/fridrichmethod/digital-twin-clinical-agent)

Trust & safety

Audits pending

Not yet audited · commit 300715d

Works with: Claude Code Claude.ai OpenAI Codex +

File contents

Digital Twin Clinical Agent

The Digital Twin Clinical Agent creates AI-powered virtual replicas of individual patients by integrating genomics, imaging, wearable data, and clinical records. These digital twins enable clinical trial simulation, treatment response prediction, and personalized therapeutic optimization, qualified by EMA and aligned with FDA guidance.

When to Use This Skill

When simulating clinical trial outcomes for drug development.
For creating patient-specific treatment response predictions.
To optimize clinical trial design and reduce sample sizes.
When predicting individual patient trajectories.
For personalized dosing and treatment selection.

Core Capabilities

Patient Digital Twin Creation: Build comprehensive patient models.
Clinical Trial Simulation: Predict trial outcomes virtually.
Treatment Response Prediction: Individualized response modeling.
Counterfactual Generation: "What-if" treatment scenarios.
Longitudinal Prediction: Forecast disease trajectories.
Trial Design Optimization: Reduce sample sizes, improve power.

Digital Twin Components

Component	Data Sources	Models
Genomic Twin	WES/WGS, RNA-seq	Mutation effects, expression
Phenotypic Twin	EHR, labs, vitals	Clinical trajectories
Imaging Twin	CT, MRI, pathology	Tumor dynamics
Behavioral Twin	Wearables, PROs	Activity, symptoms
Pharmacokinetic	Drug levels, metabolism	PK/PD models

Clinical Applications

Application	Use Case	Benefit
Trial Simulation	Virtual control arms	Reduce placebo patients
Dose Optimization	Individual PK/PD	Personalized dosing
Treatment Selection	Compare therapies	Optimal choice
Progression Prediction	Disease trajectory	Early intervention
Drop-off Prediction	Compliance forecasting	Retention improvement

Workflow

Data Collection: Gather multi-modal patient data.
Twin Construction: Build integrated patient model.
Calibration: Fit twin to individual patient data.
Validation: Compare predictions to observations.
Simulation: Run treatment scenarios.
Prediction: Generate outcome forecasts.
Output: Digital twin model, predictions, uncertainties.

Example Usage

User: "Create a digital twin for this Alzheimer's patient to simulate their response to the investigational drug and compare to placebo trajectory."

Agent Action:

python3 Skills/Clinical/Digital_Twin_Clinical_Agent/create_twin.py \
    --patient_data patient_ehr.json \
    --genomics patient_wgs.vcf \
    --imaging mri_series/ \
    --cognitive_scores mmse_history.csv \
    --biomarkers abeta_tau_nfl.csv \
    --disease alzheimers \
    --simulate_treatment drug_a \
    --compare_to placebo \
    --prediction_horizon 24_months \
    --output digital_twin_results/

Input Requirements

Data Type	Required	Purpose
Demographics	Yes	Base characteristics
Medical History	Yes	Disease context
Lab Values	Yes	Biomarker trajectories
Medications	Yes	Treatment history
Genomics	Recommended	Personalization
Imaging	Recommended	Disease state
Wearables	Optional	Real-time data
PROs	Optional	Symptom tracking

Output Components

Output	Description	Format
Digital Twin Model	Serialized patient model	.pt, .pkl
Trajectory Predictions	Future state estimates	.csv
Counterfactuals	Alternative outcomes	.csv
Uncertainty Bounds	Prediction intervals	.json
Comparison Report	Treatment vs control	.pdf
Visualization	Interactive dashboard	.html

AI/ML Components

Twin Generation:

Generative adversarial networks (ClinicalGAN)
Variational autoencoders
Large language models (DT-GPT)

Trajectory Modeling:

Recurrent neural networks
Temporal transformers
Gaussian processes

Treatment Effect:

Causal inference models
Counterfactual prediction
Potential outcomes framework

Clinical Trial Applications

Trial Phase	Digital Twin Role	Benefit
Phase I	Safety prediction	De-risk dosing
Phase II	Efficacy simulation	Go/no-go decisions
Phase III	Virtual control arm	Smaller trials
Post-marketing	Real-world outcomes	Safety monitoring

Regulatory Status

Agency	Status	Application
FDA	Guidance supportive	Acceptable with validation
EMA	Qualified	Specific use cases approved
PMDA	Under evaluation	Pilot programs

Validation Requirements

Validation Type	Method	Metric
Temporal	Hold-out future data	RMSE, calibration
External	Independent cohort	Generalization
Subgroup	Demographic splits	Fairness
Extreme	Edge cases	Robustness

Prerequisites

Python 3.10+
PyTorch, TensorFlow
Survival analysis libraries
EHR parsing tools
OMOP CDM familiarity

Related Skills

Virtual_Lab_Agent - AI research coordination
Multimodal_Radpath_Fusion_Agent - Data integration
Multi_Ancestry_PRS_Agent - Genetic risk
ctDNA_Dynamics_MRD_Agent - Disease monitoring

Disease-Specific Models

Disease	Key Endpoints	Model Maturity
Alzheimer's	ADAS-Cog, CDR	Advanced
Oncology	PFS, OS, ORR	Advanced
Cardiovascular	MACE, ejection fraction	Moderate
Diabetes	HbA1c, complications	Moderate
Multiple Sclerosis	EDSS, relapse rate	Emerging

Limitations

Limitation	Impact	Mitigation
Data Quality	Prediction accuracy	Data cleaning, imputation
Rare Events	Underrepresentation	Transfer learning
Novel Treatments	No historical data	Mechanism-based models
Individual Variation	Uncertainty	Probabilistic models

Special Considerations

Privacy: Ensure de-identification and consent
Bias: Validate across demographic groups
Interpretability: Explain predictions to clinicians
Updating: Continuously refine with new data
Uncertainty: Always quantify prediction confidence

Future Directions

Direction	Timeline	Impact
Real-time Twins	3-5 years	Continuous monitoring
Federated Twins	2-3 years	Multi-site collaboration
Causal Twins	Ongoing	True treatment effects
Regulatory Integration	5-7 years	Standard practice

Author

AI Group - Biomedical AI Platform

---name: digital-twin-clinical-agentdescription: AI-powered patient digital twin creation for clinical trial simulation, treatment outcome prediction, and personalized medicine using real-world data and multi-omics integration.---# Digital Twin Clinical AgentThe **Digital Twin Clinical Agent** creates AI-powered virtual replicas of individual patients by integrating genomics, imaging, wearable data, and clinical records. These digital twins enable clinical trial simulation, treatment response prediction, and personalized therapeutic optimization, qualified by EMA and aligned with FDA guidance.## When to Use This Skill* When simulating clinical trial outcomes for drug development.* For creating patient-specific treatment response predictions.* To optimize clinical trial design and reduce sample sizes.* When predicting individual patient trajectories.* For personalized dosing and treatment selection.## Core Capabilities1. **Patient Digital Twin Creation**: Build comprehensive patient models.2. **Clinical Trial Simulation**: Predict trial outcomes virtually.3. **Treatment Response Prediction**: Individualized response modeling.4. **Counterfactual Generation**: "What-if" treatment scenarios.5. **Longitudinal Prediction**: Forecast disease trajectories.6. **Trial Design Optimization**: Reduce sample sizes, improve power.## Digital Twin Components| Component | Data Sources | Models ||-----------|--------------|--------|| Genomic Twin | WES/WGS, RNA-seq | Mutation effects, expression || Phenotypic Twin | EHR, labs, vitals | Clinical trajectories || Imaging Twin | CT, MRI, pathology | Tumor dynamics || Behavioral Twin | Wearables, PROs | Activity, symptoms || Pharmacokinetic | Drug levels, metabolism | PK/PD models |## Clinical Applications| Application | Use Case | Benefit ||-------------|----------|---------|| Trial Simulation | Virtual control arms | Reduce placebo patients || Dose Optimization | Individual PK/PD | Personalized dosing || Treatment Selection | Compare therapies | Optimal choice || Progression Prediction | Disease trajectory | Early intervention || Drop-off Prediction | Compliance forecasting | Retention improvement |## Workflow1. **Data Collection**: Gather multi-modal patient data.2. **Twin Construction**: Build integrated patient model.3. **Calibration**: Fit twin to individual patient data.4. **Validation**: Compare predictions to observations.5. **Simulation**: Run treatment scenarios.6. **Prediction**: Generate outcome forecasts.7. **Output**: Digital twin model, predictions, uncertainties.## Example Usage**User**: "Create a digital twin for this Alzheimer's patient to simulate their response to the investigational drug and compare to placebo trajectory."**Agent Action**:```bashpython3 Skills/Clinical/Digital_Twin_Clinical_Agent/create_twin.py \ --patient_data patient_ehr.json \ --genomics patient_wgs.vcf \ --imaging mri_series/ \ --cognitive_scores mmse_history.csv \ --biomarkers abeta_tau_nfl.csv \ --disease alzheimers \ --simulate_treatment drug_a \ --compare_to placebo \ --prediction_horizon 24_months \ --output digital_twin_results/```## Input Requirements| Data Type | Required | Purpose ||-----------|----------|---------|| Demographics | Yes | Base characteristics || Medical History | Yes | Disease context || Lab Values | Yes | Biomarker trajectories || Medications | Yes | Treatment history || Genomics | Recommended | Personalization || Imaging | Recommended | Disease state || Wearables | Optional | Real-time data || PROs | Optional | Symptom tracking |## Output Components| Output | Description | Format ||--------|-------------|--------|| Digital Twin Model | Serialized patient model | .pt, .pkl || Trajectory Predictions | Future state estimates | .csv || Counterfactuals | Alternative outcomes | .csv || Uncertainty Bounds | Prediction intervals | .json || Comparison Report | Treatment vs control | .pdf || Visualization | Interactive dashboard | .html |## AI/ML Components**Twin Generation**:- Generative adversarial networks (ClinicalGAN)- Variational autoencoders- Large language models (DT-GPT)**Trajectory Modeling**:- Recurrent neural networks- Temporal transformers- Gaussian processes**Treatment Effect**:- Causal inference models- Counterfactual prediction- Potential outcomes framework## Clinical Trial Applications| Trial Phase | Digital Twin Role | Benefit ||-------------|-------------------|---------|| Phase I | Safety prediction | De-risk dosing || Phase II | Efficacy simulation | Go/no-go decisions || Phase III | Virtual control arm | Smaller trials || Post-marketing | Real-world outcomes | Safety monitoring |## Regulatory Status| Agency | Status | Application ||--------|--------|-------------|| FDA | Guidance supportive | Acceptable with validation || EMA | Qualified | Specific use cases approved || PMDA | Under evaluation | Pilot programs |## Validation Requirements| Validation Type | Method | Metric ||-----------------|--------|--------|| Temporal | Hold-out future data | RMSE, calibration || External | Independent cohort | Generalization || Subgroup | Demographic splits | Fairness || Extreme | Edge cases | Robustness |## Prerequisites* Python 3.10+* PyTorch, TensorFlow* Survival analysis libraries* EHR parsing tools* OMOP CDM familiarity## Related Skills* Virtual_Lab_Agent - AI research coordination* Multimodal_Radpath_Fusion_Agent - Data integration* Multi_Ancestry_PRS_Agent - Genetic risk* ctDNA_Dynamics_MRD_Agent - Disease monitoring## Disease-Specific Models| Disease | Key Endpoints | Model Maturity ||---------|---------------|----------------|| Alzheimer's | ADAS-Cog, CDR | Advanced || Oncology | PFS, OS, ORR | Advanced || Cardiovascular | MACE, ejection fraction | Moderate || Diabetes | HbA1c, complications | Moderate || Multiple Sclerosis | EDSS, relapse rate | Emerging |## Limitations| Limitation | Impact | Mitigation ||------------|--------|------------|| Data Quality | Prediction accuracy | Data cleaning, imputation || Rare Events | Underrepresentation | Transfer learning || Novel Treatments | No historical data | Mechanism-based models || Individual Variation | Uncertainty | Probabilistic models |## Special Considerations1. **Privacy**: Ensure de-identification and consent2. **Bias**: Validate across demographic groups3. **Interpretability**: Explain predictions to clinicians4. **Updating**: Continuously refine with new data5. **Uncertainty**: Always quantify prediction confidence## Future Directions| Direction | Timeline | Impact ||-----------|----------|--------|| Real-time Twins | 3-5 years | Continuous monitoring || Federated Twins | 2-3 years | Multi-site collaboration || Causal Twins | Ongoing | True treatment effects || Regulatory Integration | 5-7 years | Standard practice |## AuthorAI Group - Biomedical AI Platform

FridrichMethod/awesome-skills/tree/main/skills/digital-twin-clinical-agent commit 300715d8f0

Frequently asked questions

npx skillmds@latest add fridrichmethod/digital-twin-clinical-agent