A Versatile AI Agent for Rare Disease Diagnosis and Risk Gene Prioritization
Tianyu Liu, Wangjie Zheng, Rui Yang, Benny Kai Guo Loo, Hui Zhang, Jeffries Lauran, Jianlei Gu, Botao Yu, Weihao Xuan, Kexin Huang, Nan Liu, James Zou, Yonghui Jiang, Hua Xu, Hongyu Zhao · 2026
Read A Versatile AI Agent for Rare Disease Diagnosis and Risk Gene Prioritization by Tianyu Liu, Wangjie Zheng, Rui Yang, Benny Kai Guo Loo, Hui Zhang, Jeffries Lauran, Jianlei Gu, Botao Yu, Weihao Xuan, Kexin Huang, Nan Liu, James Zou, Yonghui Jiang, Hua Xu, Hongyu Zhao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-06226.txt.
Introduction
Tianyu Liu and fourteen co-authors introduce Hygieia, a multimodal AI agent for rare disease diagnosis and risk gene ranking. Rare diseases are hard to diagnose, and current workflows often mean long delays and low accuracy for patients. The system integrates phenotypic features, genetic profiles, and clinical records into one diagnostic process aimed at precision medicine.
A router based, knowledge enhanced design tailors strategy by disease category and works to limit hallucination. The agent prioritizes risk related genomic factors for rare diseases and reports confidence scores to support clinical judgment. Across diagnostic benchmarks it reached state of the art results in the authors' evaluation. In studies with clinicians at two medical schools, it beat physicians on diagnosis and helped experts work through real case records.
Catalog record
| CREATOR | Tianyu Liu, Wangjie Zheng, Rui Yang, Benny Kai Guo Loo, Hui Zhang, Jeffries Lauran, Jianlei Gu, Botao Yu, Weihao Xuan, Kexin Huang, Nan Liu, James Zou, Yonghui Jiang, Hua Xu, Hongyu Zhao |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.06226
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Versatile AI Agent for Rare Disease Diagnosis and Risk Gene Prioritization Tianyu Liu, Wangjie Zheng, Rui Yang, Benny Kai Guo Loo, Hui Zhang, Jeffries Lauran, Jianlei Gu, Botao Yu, Weihao Xuan, Kexin Huang, Nan Liu, James Zou, Yonghui Jiang, Hua Xu, Hongyu Zhao Abstract Accurate and timely diagnosis is essential for effective treatment, particularly in the context of rare diseases. However, current diagnostic workflows often lead to prolonged assessment times and low accuracy. To address these limitations, we introduce Hygieia, a multi-modal AI agent system designed to support precision disease diagnosis by integrating diverse data sources, including phenotypic features, genetic profiles, and clinical records. Hygieia features a router-based and knowledge-enhanced framework that mitigates hallucination and tailors diagnostic strategies to different disease categories. Notably, it prioritizes risk-related genomic factors for rare diseases and provides confidence scores to assist clinical decision-making. We conducted a comprehensive evaluation demonstrating that Hygieia achieves state-of-the-art performance across multiple diagnostic benchmarks. In collaboration with clinical experts from Yale School of Medicine and Duke-NUS Medical School, we further validated its practical utility by showing (1) Hygieia’s superior diagnostic performance compared to physicians with an improvement from 12%-60% and (2) its effectiveness in assisting clinicians with medical records for handling real-world cases. Our findings indicate that Hygieia not only enhances diagnostic accuracy and interpretability but also significantly reduces clinician workload, highlighting its potential as a valuable tool in clinical decision support systems. \svgsetup inkscapelatex=false keywords AI Agent, Disease Diagnosis, Rare Disease, Risk Gene Prioritization, Clinical System 1 Introduction Rare diseases are defined as conditions affecting fewer than 1 in 2,000 individuals, affecting over 300 million patients worldwide ( 41 , 28 , 31 , 16 , 48 ) . Moreover, the diagnosis and test recommendations for rare diseases are very challenging. Diagnosing rare diseases based on conventional medical methods typically takes 4 to 5 years ( 12 ) , which is known as “diagnostic odyssey”. One reason is that the phenotypes of rare diseases can sometimes be difficult to distinguish directly from common diseases. This also increases the likelihood of misdiagnosis and mistreatment by physicians ( 10 ) . To overcome the challenges mentioned above, researchers have begun collecting genotypes, phenotypes, and diagnosis plans for rare diseases to enrich our understanding of these diseases and to design a better diagnosis and treatment plan. These multi-modal datasets provide valuable research materials for disease diagnosis. Moreover, data-driven solutions, such as disease diagnosis methods based on data mining, deep learning, and advanced artificial intelligence (AI), have also garnered significant attention recently ( 19 ) . Experts can train a model using the aforementioned diagnostic data to predict diseases or prioritize disease risk genes, thereby improving diagnostic accuracy. However, such models often face shortcomings in terms of generalization and deployment ( 36 ) . To enhance the model’s versatility and accessibility for physicians, rare disease diagnostic tools based on Foundation Models (such as Large Language Models (LLMs) ( 39 ) and Visual Language Models (VLMS) ( 22 ) ) have also been developed. LLMs are pre-trained with a large-scale text corpus and can generalize into different tasks in natural language processing (NLP) with techniques of post-training. LLMs can process electronic health records (EHRs) from patients and make diagnoses accordingly ( 38 , 24 ) , and multiple LLMs with different roles can also work together as an agent system ( 40 , 11 ) . Such an AI agent can make diagnoses by simulating the real scenarios, leveraging prior knowledge, and providing recommendations and suggestions for physicians as medical AI assistants ( 23 , 25 ) . Several AI-based tools have been developed for rare disease diagnosis. For example, a general baseline for rare disease diagnosis will be prompting LLMs discussed in RareBench ( 8 ) , and RareArena ( 5 ) . Researchers also consider developing AI agents for medical usage based on techniques such as knowledge-enhanced retrieval ( 42 ) as well as multi-agent communication ( 9 , 6 ) . Although these models are interesting in design and have some clinical significance, their shortcomings are still quite evident. First, even advanced AI-based models used for rare disease diagnosis cannot distinguish between common and rare diseases, which is the basic step to avoid misdiagnosis (Supplementary Figure 1 (a)). This finding severely limits the applicability of rare disease diagnostic models. Second, due to the randomness that exists in the model training and inference, these models might not give consistent outputs based on the same input with different random seeds, which is also harmful for the trustworthy output (Supplementary Figure 1 (b)). Third, current AI methods only focus on diagnosis, but lack the necessary steps and capacities for result interpretation and discovery of important causal genes of rare diseases. Finally, current studies ( 19 , 48 , 26 ) do not directly address how these AI models can be applied in diagnostic scenarios and collaborate with physicians. Moreover, recent studies have shown that over 80% of rare diseases are influenced by genetic factors and can be passed on to the next generation ( 48 ) , but how these genetic factors are incorporated into diagnosis and how to infer disease-related genes based on clinical presentation have not been well researched in AI-based methods. Therefore, there is a critical need to design an AI model that can simultaneously diagnose different types of diseases and provide explanations for the decision. In this manuscript, we introduce Hygieia, which is an AI agent for disease diagnosis and interpretation. Our model breaks down the diagnostic process into multiple stages, first determining the disease type, then designing distinct diagnostic approaches for common and rare diseases. Diagnosing common diseases is based on prompting LLMs. Meanwhile, due to the complexity of rare disease diagnosis, our agent utilizes multiple tools (such as website searching as well as patient retrieval) to leverage prior knowledge and make a decision. We have two innovations in the design of this agent. First, to resolve the inconsistency, our agent has a verifier to monitor the outputs of the main body of Hygieia and ensure the results converge. Second, to improve the transparency of using AI agents for making clinical decisions, we implement a framework with a reasoning trajectory and confidence estimation to help users understand and trust this workflow. Hygieia can accept multiple modalities or types of data as inputs, such as phenotype information, gene information, medical history, and other information. We also provided a table in Appendix A to distinguish Hygieia versus other AI agents focusing on (rare) disease diagnosis, and the unique components (case router, confidence estimation, and multi-task capacity) of Hygieia further enhance its novelty. In general, Hygieia can interpret diagnosis results with trackable reasoning paths and prioritize disease-associated risk genes, providing more informative feedback as references to help physicians make decisions and recommendations. Hygieia can also be integrated as a skill to improve the harness of frontier AI Agents in the diagnosis of rare diseases. We invite genetic physicians with verification from Yale School of Medicine and DUKE-NUS Medical School to evaluate the contribution of Hygieia as an effective medical AI assistant, and explore new directions for the diagnosis of rare diseases at the age of AI and digital health. 2 Results