Stacey Kallish, MD, Clinical Geneticist at Penn Medicine in Philadelphia, is helping to lead a new wave of innovation at the intersection of artificial intelligence (AI) and rare disease care. With a clinical focus on lysosomal storage diseases (LSDs)—including Fabry disease, Gaucher disease, and Pompe disease—Dr. Kallish is exploring how emerging technologies can improve diagnosis, risk prediction, and long-term management for patients with complex genetic conditions.
AI is increasingly becoming a topic of discussion across healthcare, but in rare diseases, its potential impact may be especially significant. These conditions are often difficult to recognize due to their rarity, variability, and overlap with more common disorders. As a result, patients frequently experience delayed or missed diagnoses. AI offers a powerful opportunity to change that.
By training machine learning models on datasets of patients with confirmed rare diseases, researchers can teach these systems to recognize patterns that distinguish affected individuals from those without disease. These models can then be applied to large-scale datasets, such as electronic health records (EHRs), to flag patients who may otherwise go undiagnosed.
Dr. Kallish explains that one of AI’s greatest strengths lies in its ability to synthesize vast amounts of information. “These models can evaluate countless small data points simultaneously—far more than we can realistically process as clinicians,” she notes. This allows AI to detect subtle signals that may not be immediately apparent, particularly in early or atypical cases.
Beyond diagnosis, AI is also being investigated as a tool for predicting disease progression and guiding treatment decisions. In conditions like Fabry disease, clinicians must often determine when to initiate therapy and how to assess a patient’s risk for complications such as stroke or organ damage. AI-driven models have the potential to analyze longitudinal patient data and generate individualized risk profiles, offering more precise and personalized insights.
However, while these applications are promising, Dr. Kallish emphasizes that the field is still evolving. Diagnostic use cases—such as identifying patients through EHR screening or interpreting genetic variants—are likely to reach clinical utility sooner. More complex applications, like optimizing treatment timing or selecting the most appropriate therapy for an individual patient, will require additional validation and real-world evidence.
Importantly, AI is not intended to replace physicians. Instead, it is best understood as a complementary tool—one that can enhance clinical decision-making and improve access to specialized expertise. This is particularly evident in the field of medical imaging.
Studies have shown that AI models can perform at a level comparable to expert radiologists in interpreting imaging studies such as brain MRIs. In some cases, these models even outperform general radiologists. But rather than viewing this as a threat, Dr. Kallish sees it as an opportunity for collaboration.
“The goal is partnership,” she explains. “AI can help us get to better answers, especially in settings where specialized expertise may not be readily available.” For patients in community hospitals or underserved areas, this could mean more accurate diagnoses and improved care without the need for referral to major academic centers.
Another area where AI may have a transformative impact is biomarker development. Currently, clinicians rely on a combination of enzyme assays, molecular testing, and established biomarkers to diagnose and monitor LSDs. AI has the potential to identify new biomarkers that are more sensitive, more specific, and potentially tailored to individual patients. These insights could help clinicians better determine when to initiate treatment, assess disease progression, and evaluate therapeutic response.
Some AI-driven tools are already making their way into clinical practice. One example is FDrisk, a publicly available platform that uses machine learning to estimate the likelihood of Fabry disease based on a series of clinical inputs. By answering a set of targeted questions, clinicians can generate a dynamic risk score and receive guidance on next steps, including whether to pursue diagnostic testing.
At a broader level, researchers are also leveraging AI to analyze patient journeys within healthcare systems. By studying the clinical trajectories of patients with known rare diseases, machine learning models can identify similar patterns in other patients’ records. This approach has already demonstrated success in detecting conditions such as Rett syndrome and Lowe syndrome, highlighting the potential for earlier identification and intervention.
3
Rare Endocrine Exchange: Why This Conversation Matters
CheckRare August 16, 2026 5:50 am