This article, the 14th installment of Dr. Stanley Terlecky's "Terlecky's Corner," was published June 18, 2025.
While defining rare diseases1 consistently across medical and scientific communities presents challenges, what is certain is that they affect millions of people worldwide. The suffering is enormous - and the resultant cost to patients, their families, and society more broadly - is incalculable. Conspiring against rare disease study is the fact that only a limited number of patients would benefit from the drugs developed. This economic reality has led the pharmaceutical industry to often neglect - or “orphan” - these diseases. Despite attempts by patient advocates, the Orphan Drug Act of 1983 (which created financial incentives to investigate rare diseases), and the tremendous cellular and molecular advances made in biomedicine over recent decades, rare diseases remained poorly diagnosed, often devoid of treatment options and effective prevention strategies. This brings us to today’s seismic shifts in technology development, including artificial intelligence (AI), which may finally bring relief to the long-suffering masses across the globe.
One promising avenue, significantly boosted by AI, is drug repurposing - also referred to as drug repositioning - which already exists. Consider sildenafil, originally developed to lower blood pressure, which was later found (as Viagra®) to also address erectile dysfunction. The idea is to reexamine an approved drug - looking for “off target” drug effects - that is, beneficial secondary activities which may directly impact the molecular mechanisms associated with rare diseases. Importantly, this is done not on a small scale with a few drugs - but rather through use of powerful (even quantum) computers. Machine learning can analyze approved drugs’ myriad effects, with AI powering the assessment of potential applicability to rare disease manifestations. It represents a potential new frontier in modern drug development. The drugs are out there - their effects known - what is needed is a mechanism to bring together the drugs’ actions and the patients’ clinical needs. More formally restated: a translation framework that systematically connects existing pharmaceutical properties with unmet therapeutic needs. What is so exciting is that the time is right for this - indeed, the time is now! We know drugs often have multitarget/broad spectrum effects and we know their actions are often outside their primary mechanism of action. AI has been extensively trained - algorithms are emerging and drugs are being appropriately identified for clinical use. Consider the amazing work of Dr. Marinka Zitnik2 at Harvard’s Department of Biomedical Informatics. She has leveraged “geometric deep learning and human centered AI” to develop TxGNN - “a model for identifying therapeutic opportunities for disease with limited treatment options and minimal molecular understanding”.3
TxGNN Explorer, the visual interface for the Huang, Et al. paper, Zero-shot drug repurposing with geometric deep learning and clinician centered design, which propose TxGNN for identifying therapeutic opportunities for diseases with limited treatment options and minimal molecular understanding.
A beautiful story documenting the power of such an approach is found in the work4 of David Fajgenbaum, MD - an investigator at the Perelman School of Medicine at the University of Pennsylvania. Dr. Fajgenbaum and colleagues showed that adalimumab (know commercially as Humira®), a tumor necrosis factor-alpha (TNF-alpha) inhibitor - and approved for treatment of adult rheumatoid arthritis, juvenile idiopathic arthritis, ankylosing spondylitis, psoriasis, Crohn’s disease, and ulcerative colitis - proved to be a life-saving therapy for idiopathic multicentric Castleman’s disease (iMCD). iMCD is, by every measure, a rare disease - one whose presence is life-threatening and for which few therapeutic options exist. An AI-powered algorithm suggested adalimumab could be effective in iMCD - via its ability to thwart action of TNF-alpha. Patients with severe iMCD do express high levels of TNF-alpha (among other pro-inflammatory cytokines). The patient was treated with the anti-TNF drug, and the intervention yielded amazing results: the patient has been disease-free for two years.5
These AI-based drug repurposing algorithms have the potential to revolutionize drug discovery - and for the first time - begin to address the historical challenges associated with rare diseases. Importantly, it could be envisioned that both TxGNN and the AI-platform employed in the iMCD study could not only impact rare diseases with limited understanding of molecular mechanisms, but could also address more common pathologies where current therapeutics are less than completely effective or possess protean side-effects. The drugs exist - their safety and efficacy profiles are documented; there is no need to develop new drugs at great expense. The approach is rational - and based on data already obtained and analyses already completed. Once again, the time is now - and for millions of patients with disease around the world - it could not come soon enough.
SRT - June 2025
References:
0 Comments.