‹ Back

Genomics, I’m Back — with GenAI + LHS

 

ELHS Newsletter 2025-11-15

(Paving the Way for Global Health Equity with Generative AI and Learning Health Systems)

Subscribe | All Issues

 

Dear Friends,

 

October was a busy month. Together with my collaborators at Stanford and Harvard, I presented our vision paper at the Chen Institute Symposium 2025 for AI Accelerated Science in San Francisco, proposing a new framework for Open Clinical AI Science. Our aim is to accelerate the generation of clinical evidence for GenAI effectiveness by 10×, leveraging the emerging convergence of Generative AI and Learning Health Systems (LHS).

At the invitation of the Dean of the Tsinghua University Medical School, I also delivered a keynote address on this convergence at the Tsinghua Health AI Summit in Beijing. I first explored the interplay between GenAI and LHS in a review paper last year, focusing on the democratization of GenAI in healthcare. Today, I am more convinced than ever that combining GenAI and LHS holds the promise to rescue the highly anticipated healthcare AI industry from a potential crash driven by an underestimation of the time required to generate clinical GenAI evidence across all diseases.

At the same time, the U.S. National Academies convened an important workshop, Exploring Applications of AI in Genomics and Precision Health, where national experts discussed current and future opportunities for AI across the spectrum from translational research to clinical care. Dr. Nephi Walton alerted clinicians that, according to a study presented at ASHG 2025, LLMs may outperform clinical geneticists in rare disease diagnosis. He also shared several cases in which patients identified their own genetic diseases using LLMs before ever reaching a clinical geneticist. He emphasized that scaling genomics requires streamlining the delivery of both genomics and AI to the front lines of care. Dr. Melissa Haendel highlighted that more than 10,000 rare diseases have been identified across multiple health systems, and she described numerous roadblocks that hinder timely diagnosis for patients with rare diseases — insights grounded in the community perspective to which I also contributed.

The NASEM workshop brought back vivid memories of my own journey on the genomics battlefield over the years. I began pursuing my dream of a genomic health revolution in the late 1990s. At the height of the first genomics industry wave in 2000, I led the consumer software group at Hyseq Inc., where we developed what was likely the first direct-to-consumer genomic sequence information product line in history. But when the genomics industry soon crashed, I came to a stark realization: before genomics can truly transform healthcare, we must first solve its hardest challenge—generating clinical evidence demonstrating the effectiveness of genomic information for diagnosing and treating all diseases.

Without such evidence, even achieving the once “crazy” goal of the $1,000 genome would not be enough to revolutionize healthcare. Around 2012, when the U.S. National Academy of Medicine articulated its vision for Learning Health Systems to address inefficiencies in clinical evidence generation and dissemination, I saw a promising path forward. Yet for more than a decade, no technology existed to enable LHS implementation at scale and at low cost.

A breakthrough came in 2022, when I successfully generated synthetic patient data to simulate what I called an “LHS unit” for lung cancer risk prediction. Then, in 2023, the arrival of ChatGPT changed everything. My benchmarking studies demonstrated nearly 80% prediction accuracy across about 200 diseases, convincing me that it is now feasible to build cost-effective LHS units in virtually any clinical setting. These units can continuously learn and refine genomic algorithms from real-world data using Generative AI, thereby accelerating the development of clinical genomics evidence across all diseases in parallel.

How significant is this breakthrough? Time will tell. For now, I’m back—back to genomics, and back with GenAI and LHS to finish the work I began a quarter-century ago. I’ll be talking more with my copilot, ChatGPT, about the next phase of this genomics battle.

Notably, the State of AI Report 2025 from Nathan Benaich and Air Street Capital contains a wealth of valuable information about the rapid progress of LLM, GenAI, and AGI.

 

Warm regards,
AJ

AJ Chen, PhD
Founder and PI, ELHS Institute
Silicon Valley, USA
ELHS Newsletters: https://elhsi.org/Newsletters
ELHS GenAI Platform: https://elhsi.com

 

~

 

From Page Mill

(Recent papers, news, and events showcasing the progress of GenAI and LHS) 

Angus DC, Khera R, Lieu T, et al. AI, Health, and Health Care Today and Tomorrow: The JAMA Summit Report on Artificial Intelligence. JAMA. Published online October 13, 2025.

 

[2025/10] Ensuring AI is deployed equitably and in a manner that improves health outcomes or, if improving efficiency of health care delivery, does so safely, requires progress in 4 areas. First, multistakeholder engagement throughout the total product life cycle is needed. This effort would include greater partnership of end users with developers in initial tool creation and greater partnership of developers, regulators, and health care systems in the evaluation of tools as they are deployed. Second, measurement tools for evaluation and monitoring should be developed and disseminated. Beyond proposed monitoring and certification initiatives, this will require new methods and expertise to allow health care systems to conduct or participate in rapid, efficient, and robust evaluations of effectiveness. The third priority is creation of a nationally representative data infrastructure and learning environment to support the generation of generalizable knowledge about health effects of AI tools across different settings. Fourth, an incentive structure should be promoted, using market forces and policy levers, to drive these changes.

 

Chen H, Zeng D, Qin Y, et al. Large language models and global health equity: a roadmap for equitable adoption in LMICs. Lancet Reg Health West Pac. 2025 Oct 18;63:101707.

 

[2025/10] Large language models (LLMs) have been proposed to address global health inequity by providing accessible and high-quality health care, particularly in low- and middle-income countries (LMICs). However, despite the early enthusiasm following the release of GPT, development and deployment of LLMs have remained heavily concentrated in high-income countries (HIC), raising concerns that such technology may worsen existing health disparities instead of alleviating them. The most recent LLMs, which include features such as lower cost, and open-source framework, show promise in rebalancing LLMs' benefits worldwide. In this viewpoint, we examine the current challenges and imbalance in LLM deployment across global regions, identify the key barriers to adoption in LMICs, assess current LLMs' advances and the new opportunities they bring to global health equity. We also propose a five-dimensional roadmap—focusing on people, products, platforms, processes, and policies—to advance LLMs' equitable adoption in LMIC and improve inclusive progress in global health.

 

Mahajan, A., LaChance, A.H., Rodman, A. et al. Artificial intelligence for autoimmune diseases. npj Digit. Med. 8, 628 (2025).

 

[2025/10] Emerging evidence suggests generative artificial intelligence (AI) may offer potential for autoimmune and rheumatic disease care, moving beyond traditional narrow AI applications to produce contextualized clinical content to support a wide spectrum of medical tasks. This article explores generative AI applications across autoimmune and rheumatologic clinical care, research, and administrative domains. However, significant implementation challenges remain, including clinical validation, model interpretability, data integration complexities, and evolving regulatory frameworks.

 

Jin, Q., Wang, Z., Yang, Y. et al. AgentMD: Empowering language agents for risk prediction with large-scale clinical tool learning. Nat Commun 16, 9377 (2025).

 

[2025/10] We introduce AgentMD, a novel language agent capable of curating and applying clinical calculators across various clinical contexts. As a tool builder, AgentMD first uses PubMed to curate a diverse set of 2,164 executable clinical calculators with over 85% accuracy for quality checks and over 90% pass rate for unit tests. As a tool user, AgentMD autonomously selects and applies the relevant clinical calculators. Our evaluations show that AgentMD significantly outperforms GPT-4 for risk prediction (87.7% vs. 40.9% in accuracy). Results on 698 real-world emergency department notes confirm that AgentMD accurately computes medical risks at an individual level. Moreover, AgentMD can provide population-level insights for institutional risk management. Our study illustrates the capabilities of language agents to curate and utilize clinical calculators for both individual patient care and at-scale healthcare analytics.

 

Jarchow, H., Bobrowski, C., Falk, S. et al. Benchmarking large language models for personalized, biomarker-based health intervention recommendations. npj Digit. Med. 8, 631 (2025).

 

[2025/10] We extended the BioChatter framework to benchmark LLMs’ ability to generate personalized longevity intervention recommendations based on biomarker profiles while adhering to key medical validation requirements. Using 25 individual profiles across three different age groups, we generated 1000 diverse test cases covering interventions such as caloric restriction, fasting and supplements. Evaluating 56000 model responses via an LLM-as-a-Judge system with clinician validated ground truths, we found that proprietary models outperformed open-source models especially in comprehensiveness. However, even with Retrieval-Augmented Generation (RAG), all models exhibited limitations in addressing key medical validation requirements, prompt stability, and handling age-related biases. Our findings highlight limited suitability of LLMs for unsupervised longevity intervention recommendations. 

 

Makarov, N., Bordukova, M., Quengdaeng, P. et al. Large language models forecast patient health trajectories enabling digital twins. npj Digit. Med. 8, 588 (2025).

 

[2025/10] We developed the Digital Twin—Generative Pretrained Transformer (DT-GPT), extending LLM-based forecasting solutions to clinical trajectory prediction. DT-GPT leverages electronic health records without requiring data imputation or normalization and overcomes real-world data challenges such as missingness, noise, and limited sample sizes. Benchmarking on non-small cell lung cancer, intensive care unit, and Alzheimer’s disease datasets, DT-GPT outperformed state-of-the-art machine learning models, reducing the scaled mean absolute error by 3.4%, 1.3% and 1.8%, respectively. It maintained distributions and cross-correlations of clinical variables, and demonstrated explainability through a human-interpretable interface.

 

Tripathi, A., Waqas, A., Schabath, M.B. et al. HONeYBEE: enabling scalable multimodal AI in oncology through foundation model-driven embeddings. npj Digit. Med. 8, 622 (2025).

 

[2025/10] Evaluated on 11,400+ patients across 33 cancer types from The Cancer Genome Atlas (TCGA), clinical embeddings showed the strongest single-modality performance with 98.5% classification accuracy and 96.4% precision@10 in patient retrieval. They also achieved the highest survival prediction concordance indices across most cancer types. Multimodal fusion provided complementary benefits for specific cancers, improving overall survival prediction beyond clinical features alone. Comparative evaluation of four large language models revealed that general-purpose models like Qwen3 outperformed specialized medical models for clinical text representation, though task-specific fine-tuning improved performance on heterogeneous data such as pathology reports.

 

Boyer, L., Fernandes, S., Auquier, P. et al. Reimagining patient-reported outcomes in the age of generative AI. npj Digit. Med. 8, 624 (2025).

 

[2025//10] Despite decades of psychometric and digital innovation, PROs remain conceptually limited and underused in both clinical practice and AI models. Rooted in top-down, predefined instruments and assumptions of unidimensionality, traditional PROs struggle to capture the fluctuating and multidimensional nature of lived health experiences. In contrast, generative AI supports bottom-up, narrative-based approaches that process language in a flexible and context-aware way. Our viewpoint supports two distinct directions: one that refines current psychometric models through generative artificial intelligence integration, and another that embraces a more disruptive shift toward language-native tools capable of synthesising patient narratives. 

 

~

 

(Source: Dr. Nephi Walton presentation on NASEM precision medicine workshop)

 

 

In the Eyes of GPTs

(AJ in conversation with ChatGPT to understand how machines view human healthcare)

 

  Let’s discuss the previous phases of the genomics industry battle and how to win the new phase. What do you know about the genomics industry battle and its earlier phases?

I’ve simplified the previous phases into four categories:

1. The Human Genome Project (HGP)
2. The genomics industry boom and bust
3. The NGS arms race
4. The rise of direct-to-consumer (DTC) genomics

The current Phase 5 focuses on how GenAI can scale up clinical evidence generation.

 

 ChatGPT

The Genomics Industry Battle — What the Previous Phases Looked Like

The genomics industry didn’t progress in a straight line. It unfolded in distinct waves, each with its own ambitions, breakthroughs, hype cycles, crashes, and lessons. These phases shaped the landscape you re-enter now with GenAI + LHS.

Phase 1 — Human Genome Project (HGP)

Battle Theme: “Map the genome so the revolution can begin.”

Accomplishments

• Global scientific mobilization
• Reference genome completed
• Created the foundation for genomic medicine

Limitation (the invisible force emerges):
Human whole genome sequence led humans enter into the BioInfo age where we have the complete blueprint of our species. However, mapping the genome did not generate clinical evidence.

Phase 2 — Genomics Industry Boom and Bust

Battle Theme: “Turn genomics into products before evidence exists.”

Accomplishments

• High innovation and entrepreneurial energy
• Early genomic diagnostics and DTC prototypes
• Tremendous public excitement

Crash causes

• Evidence gap
• Regulatory uncertainty
• Market readiness overestimated

This was the first time the invisible force crushed an entire genomics industry. The top four public genomics companies—including Hyseq, where I worked at the time—disappeared forever, wiping out billions of dollars in investment. The crash of this first genomics industry wave taught me a critical insight: genomics cannot transform healthcare without scalable clinical evidence generation.

Phase 3 — NGS Arms Race

Battle Theme: “Drive sequencing cost to zero.”

Accomplishments

• $100M → $1,000 genome
• Explosion of sequencing technologies
• Adoption in oncology and rare diseases

Limitation
Even with cheap genomes, clinical utility remained limited.
The invisible force adapted but stayed undefeated.

Phase 4 — Rise of DTC Genomics

Battle Theme: “Scale genomics through consumers.”

Accomplishments

• Tens of millions of genotypes
• Expanded public awareness
• Accelerated variant databases

Limitation
DTC genomics rarely translated into clinical action. 23andMe filed bankruptcy.
Again, evidence generation remained slow, fragmented, and expensive.

Phase 5 — GenAI Era: The First Chance to Defeat the Invisible Force

Battle Theme: “Use GenAI + LHS to scale clinical evidence generation by 10×.”

Here is where your work becomes strategic, visionary, and timely.

GenAI brings:

• High diagnostic accuracy across almost all diseases
• Ability to reason over genomics + clinical + environmental data
• Synthetic data generation for safe prototyping, data sharing, and copilot training
• Natural-language interfaces for clinicians and patients

Learning Health Systems (LHS) bring:

• Continuous real-world data integration
• Evidence generation within routine practice
• Rapid hypothesis testing and iteration
• Clinical-grade feedback loops

Together, they form a new Open Clinical AI Science ecosystem.

This is the first framework in history that can potentially neutralize the invisible force.

 

 Please lay out a battle plan for the genomics industry to win against the invisible (dark) force. Our proposed new framework—based on the convergence of GenAI and Learning Health Systems (LHS) in clinical science—represents a promising approach.

 

 ChatGPT

The Invisible Force: What Is It?

Define it explicitly so the battle plan is clear:

The Invisible (Dark) Force

The structural barriers that make clinical evidence expensive, slow, and fragmented:

• Long trial timelines
• Fragmented EHR systems
• Lack of scalable clinical validation
• Limited adoption of genomic algorithms
• Clinician workflow friction
• Data privacy constraints
• Insufficient learning loops in real-world practice

For 25 years, genomics has been fighting this force without proper weapons.

GenAI + LHS changes that.

 

THE BATTLE PLAN: How Genomics Can Win Phase 5

1. Build LHS Units for Every Disease

Treat each disease as a mini learning health system unit that:

• Continuously collects real-world data
• Evaluates genomic signals
• Refines models over time
• Generates higher-resolution evidence every day

Objective: Make evidence generation automatic.

2. Deploy GenAI as the Universal Genomics Interpreter

Use GenAI to:

• Integrate genomic + clinical + imaging + environmental data
• Predict disease trajectories
• Identify pathogenic variants
• Provide risk explanations for clinicians
• Suggest next diagnostic steps

Objective: Turn genomic information into clinical action at scale.

3. Use Synthetic Patient Data as the Training Ground

Before touching real patients:

• Build synthetic populations
• Train disease-specific LLMs
• Validate genomic algorithms
• Test variant–phenotype relationships
• Simulate clinical outcomes

Objective: Shorten the evidence pipeline from years → months.

4. Embed GenAI + LHS into Real Clinical Workflows

Inside EHR systems:

• Trigger genomic alerts
• Generate diagnostic suggestions
• Support decision-making
• Produce evidence while patients are being treated

Objective: Convert every clinic into a real-time evidence generator.

5. Connect LHS Units into Cross-Institutional Clinical Research Networks (CRNs)

Build parallel learning systems across institutions:

• Harmonized FHIR data
• Shared synthetic patient data, fine-tuned LLMs, and evidence pipelines
• Comparable genomic endpoints

Objective: Scale evidence generation 10× nationally → 100× globally.

7. Establish Open Clinical AI Science (OCAIS)

A new scientific framework that:

• Unifies GenAI + LHS
• Ensures reproducibility
• Enables parallel evidence generation
• Provides an open ecosystem for genomic algorithms

Objective: Create the infrastructure that dismantles the invisible force permanently.

8. Shift the Industry Narrative

Stop asking:

“How do we make genomics cheaper?”

Start asking:

“How do we make clinical evidence generation 10× faster, 10× cheaper, and 10× more scalable?”

This is the real revolution.

The Victory Condition: When Do We Know We’ve Won?

We win the genomics battle when:

• Every genomic algorithm has continuously updating, real-world evidence
• Clinical genomics is embedded in daily care
• Evidence generation is automatic and parallel
• Doctors use GenAI copilots to validate genomic signals
• Each clinic operates as an LHS unit
• Genomics improves patient outcomes across all diseases in real time

This will fulfill the original promise of the Human Genome Project— not because sequencing got cheaper, but because clinical evidence generation and dissemination finally scales.

 

~

 

 

🤖 GenAI-ELHS Solution Empowering physicians to accelerate the responsible use of GenAI in healthcare through three growth stages: 🚀 1. GenAI Quick Start – Access GenAI agents preclinically validated with synthetic data to enhance routine clinical decision-making. 🛠️ 2. LLM Customization – Fine-tune LLMs with your own data within ELHS units to adapt GenAI to your clinical setting. 🌍 3. Broader Impact – Scale your ELHS units into research networks so more physicians can use your LLMs, collaboratively improving care quality and outcomes for larger populations.

🧠 Meta Open-Source AI Creating a positive future by making innovation available to all and enabling advancements that can enhance our lives and spur economic growth. Open-source AI provides tools for more people to build awesome things and expand our horizons.

 

~

 

🔹 ELHS Institute Relevant Resources 🔹

🤖 ELHS GenAI Copilot Platform

🌐 ELHS Institute Website

📰 ELHS Newsletters

▶️ ELHS Videos

👉 For Clinical AI technology support, contact us at support@elhsi.org 📩

 

~ the end ~