Rare diseases, which affect more than 350 million people worldwide, are often difficult to diagnose and even more difficult to treat. Many patients wait five years or more for a correct diagnosis and are misdiagnosed at least once and over 90% of conditions still lack effective therapies.

People living with rare diseases and their caregivers often lack recognition and effective tools to manage their conditions—despite being the true experts in what it means to live with a rare disease.

Artificial intelligence (AI) can be an important ally for improving the lives of people with rare diseases, their physicians, researchers and drug makers in the rare disease space:

· Helping to improve diagnosis

· Helping to find better treatments

· Empowering individuals living with rare diseases to better understand and manage their disease.

Connecting those important areas, AI can thus help put rare disease patients in the center. As AI technology matures, rare disease patients could be among the first to benefit from a smarter, faster, and more patient-centric era of drug development.

Finding invisible patterns – How AI can help shorten the diagnostic odyssey

Rare disease patients often wait years for a diagnosis. In addition to being a significant burden to patients’ well-being, low diagnostic rates also impact research and development in the rare disease space and challenge the success of marketed orphan drugs.

A recent study published in Nature’s European Journal of Human Genetics analyzed data from 6,507 participants across 41 countries and 1,675 rare diseases, highlighting the prolonged and often frustrating diagnostic journey faced by people living with rare diseases. While initiatives like the European Rare 2030 foresight study and IRDiRC recommend diagnosis within six to twelve months, the average diagnostic timeline in Europe stands at 4.7 years. Half of the patients waited at least nine months, and a quarter waited over five years for a confirmed diagnosis. Nearly three-quarters of participants experienced at least one misdiagnosis, and over 20% consulted more than eight healthcare professionals before reaching the correct diagnosis.

The main causes of delay stem from systemic healthcare issues—accounting for 90% of diagnostic lag—such as limited physician awareness of rare diseases, long referral times, and inadequate access to appropriate diagnostic tools. Given the low prevalence of individual rare diseases, recognizing relevant patterns can be particularly difficult for general practitioners and even specialists.

AI can be an important ally for improving rare disease diagnoses, by detecting patterns that might be missed by the human eye.

At UC San Francisco and UCLA, researchers developed a machine learning algorithm that combs through electronic health records to spot patterns indicative of acute hepatic porphyria, a rare disease with gastrointestinal symptoms that resemble endometriosis, appendicitis, and other disorders. The AI successfully flagged patients who might have this condition, potentially shortening a diagnostic odyssey that can otherwise take up to 15 years.

Notably, about 80% of rare diseases have a genetic origin and many have unique phenotypic patterns, such as distinct facial features that occur in around 40% of rare disease patients and unique sets of symptoms. AI-enabled next-generation sequencing (NGS) analysis and next-generation phenotyping (NGP) can help uncover those subtle pattern from complex datasets to inform diagnosis

The improved accuracy, cost-efficiency and comprehensiveness of sequencing methods have generated a vast pool of genomic data, including variants across ~4,000 genes linked to ~6,500 rare diseases and their annotated phenotypes. Yet pinpointing disease-causing variants in these large datasets remains challenging, especially as some biobanks lack sufficient evidence for pathogenicity. AI and machine learning tools—such as those developed by Berlin-based Lucid Genomics and U.S.-based Genomenon—can help automate variant interpretation and speed up diagnosis.

NGP tools are increasingly helping to close diagnostic gaps by using AI to analyze medical images and detect patterns more objectively than humans. This enables clinicians to reach faster and more accurate diagnoses, based on facial recognition and other image types, such as skeletal radiographs and retinal scans. Since its clinical utility was first demonstrated in 2014, the accuracy and applicability of facial recognition tools, such as Face2Gene and GestaltMatcher have improved significantly with deep learning advancements, allowing clinicians to analyze patient photographs via app and getting predictions for diagnosis and suggestions for genetic tests.

Still, like all AI applications, reliability depends on the quality and diversity of training data. Biases in genomic and phenotypic datasets – such as focus on patients from Western populations - must be addressed to ensure accurate, equitable diagnostics.

Grasping the disease root – how AI can aid faster and smarter drug discovery

A deep understanding of disease biology is essential for developing effective treatments, as it enables the identification of targets that are both addressable and relevant to the disease. However, for many rare diseases, both the underlying causes and their connection to clinical symptoms remain poorly understood. This lack of clarity poses a significant challenge for identifying targeted therapies.

Traditional drug discovery models—often requiring over a decade of development and facing failure rates around 90%—are poorly suited to rare diseases. In contrast, innovative approaches that incorporate disease-specific insights and data-driven methods hold promise for accelerating development timelines and reducing risk.

Notably, several AI-driven biotech companies like Recursion, Insilico Medicine, and Verge Genomics have built their discovery platforms around algorithmic target and molecule identification. Around half of their candidate programs focus on rare diseases like ALS (Amyotrophic Lateral Sclerosis), idiopathic pulmonary fibrosis (IPF), and rare cancers —a proportion that aligns closely with the share of rare disease treatments entering the broader drug development pipeline.

In addition to AI-driven approaches for novel target and molecule discovery, some innovators are focusing on accelerating rare disease treatment through drug repurposing. Healx, for example, uses its Healnet platform to integrate diverse rare disease data into a knowledge graph that reveals promising disease-compound connections. In 2024, Healx’s AI-identified compound HLX-1502 (an oral drug) received FDA clearance to begin a Phase II trial for neurofibromatosis type 1 (a rare genetic tumor disorder). Similarly, the nonprofit Every Cure is undertaking a three-year initiative to identify repurposable drugs for rare diseases using AI. The project recently received $48 million in funding from the U.S. Advanced Research Projects Agency for Health (ARPA-H), underscoring growing institutional support for AI-enabled repurposing strategies.

Clinical trials for rare diseases face a distinct set of challenges that make traditional approaches difficult to implement. Patient populations are often small, geographically dispersed, and heterogeneous. In many cases, standard biomarkers or trial endpoints are unavailable, and the use of placebo groups may raise ethical concerns—particularly in diseases with severe or life-threatening outcomes.

AI technologies are emerging as powerful tools to address these hurdles. Companies like IPM.ai and Saama use AI to identify eligible patients from real-world data and optimize trial logistics, helping researchers reach the right participants more efficiently. Meanwhile, Unlearn.ai is advancing the use of digital twins—AI-generated virtual control groups that can simulate placebo arms, reducing the number of participants needed and addressing ethical concerns around placebo use.

While AI holds significant promise, its application in clinical trials for rare diseases remains less mature compared to its use in molecule discovery and diagnosis. However, the recent surge in generative AI and the rise of user-friendly large language models are expected to drive the development of new tools that will make rare disease trials more efficient and centered around patient needs.

Patients to the center - how AI can help empower people with rare diseases and their families

AI tools can also directly help patients to understand their disease, find specialists, clinical trials and treatments and connect with peers who share similar stories.

Given that 90% of rare diseases still lack specific treatments, one option for rare disease patients is participating in clinical trials which test experimental medicines. But finding a trial that matches the disease can be challenging. Researchers at NIH created an AI tool called TrialGPT that uses a large language model to match patients with relevant clinical trials. Given a patient’s medical summary, TrialGPT automatically scans the entire ClinicalTrials.gov database and produces a list of trials the patient is likely eligible for, excluding trials that don’t fit. Importantly, the AI also provides a plain-language explanation of how each recommended trial’s criteria apply to the patients. Similarly, the Peearz mobile app was designed to help rare disease patients find and enroll in clinical trials. It leverages sophisticated AI algorithms to match a patient’s specific condition and medical profile with active trials seeking participants.

AI can also help rare disease patients and their caretakers and families in finding and building community. AI-powered Health platforms like Somebody To Talk To (STTT) and Citizen Health, which recently partnered with the Chan Zuckerberg Initiative’s Rare As One project, help rare disease patients connect with a community of peers and get a better understanding of their disease. For example, STTT uses artificial intelligence to analyze the discussions from patient support group sessions and identify common themes or insights across those conversations.

Another area in which AI-powered apps can help rare disease patients and their families is with disease management apps such as HealthStoryAI, which help capture symptoms, diet, medication and other factors and apps which help foster therapy adherence for the often complex medication schedules for rare diseases

While these initiatives have the potential to help rare disease patients and their families become more empowered, connected and knowledgeable about their own disease, this must be carefully weighed against questions of data privacy and transparency of AI applications.

Conclusion

AI won’t solve every challenge in rare disease drug development—but it can shift the odds. From diagnosing conditions faster to finding novel treatments and redesigning clinical trials, AI has the potential to make rare diseases less rare in the eyes of medicine. At the same time it can help patients being more informed about their disease and take a more active part in managing treatment.