AI Helps Solve Rare Disease Mysteries

In a significant stride for medical diagnostics, a research collaboration between OpenAI and the Manton Center for Orphan Disease Research at Boston Children's Hospital and Harvard has demonstrated how artificial intelligence can help unravel complex medical mysteries. Rare diseases affect hundreds of millions of families worldwide, with the diagnostic journey often taking six to seven years. This new AI-driven workflow has shown promise in shortening that path, successfully identifying 18 rare disease diagnoses in 376 cases studied.

The Diagnostic Odyssey

The conversation featured Stav Rones, a research participant who shared his personal experience with a rare condition. He described the years-long struggle to get a diagnosis, initially misdiagnosed before undergoing genome sequencing. The clarity of a diagnosis, he explained, was life-changing, enabling him to connect with a community, understand his prognosis, and plan for his family's future.

The full discussion can be found on OpenAI Youtube's YouTube channel.

Dr. Alan Begs, Director of the Manton Center, highlighted that experiences like Stav's are all too common, referring to it as a "diagnostic odyssey." He explained the sheer complexity of the human genome, with its billions of bases encoding tens of thousands of genes, and the thousands of genes known to be associated with disease when abnormal. Historically, the ability to test genes was limited, making diagnosis a painstaking process.

Dr. Katherine Brownstein, Scientific Director of the Manton Center's Gene Discovery Corps, elaborated on the bottleneck of time. She described the meticulous process geneticists undertake, sifting through variants, cross-referencing databases, and analyzing complex symptoms. "There's a huge amount of work that goes into looking at each and every variant," she noted, emphasizing how an AI model could significantly accelerate this process.

AI's Role in Accelerating Discovery

Suyash Shringarpure, a Machine Learning Researcher at OpenAI, explained the team's approach. They wondered if AI models could expedite the diagnostic process by performing literature searches, identifying hypotheses, and presenting a curated list to experts. The core idea was to use AI to "accelerate that by having the AI models do the literature search, identify hypothesis and present that list to a person to prioritize."

The team initially tested the AI on diagnosed cases to validate its accuracy before applying it to unsolved cases. Shringarpure noted that the AI had to be prompted carefully to avoid common mistakes in analyzing complex genetic data, such as requiring high sequencing depth to confirm variant accuracy. Once refined, the AI could identify cases correctly up to 80-90% of the time, making it a valuable tool for human analysts.

Uncovering New Leads and Potential Cures

In the study, the AI workflow surfaced evidence that led to 18 new diagnoses, some of which were surprising. Dr. Brownstein shared an example where the AI nominated a gene, S1PR1, for a complex case involving vitiligo, transposition of great vessels, and pulmonary hypertension. This suggestion, initially met with skepticism, led to the discovery of a 26-year-old paper linking S1PR1 to vitiligo, and further research is now underway.

Dr. Begs highlighted the potential for AI to synthesize information and propose novel hypotheses, which is a critical next step in medical research. He also emphasized that AI tools do not replace human diagnosticians but rather make their process faster and more efficient, allowing them to focus on the most challenging cases.

The Future of AI in Healthcare

The speakers expressed optimism about the future of AI in medicine. Shringarpure envisions AI models reanalyzing cases as new research emerges, surfacing relevant findings to researchers. Dr. Begs sees AI as a way to make diagnoses more efficient and to potentially flag new findings in previously unsolved cases.

Dr. Brownstein believes they are "just getting started" and envisions a future where undiagnosed cases can be run through AI tools, matched with diagnoses, and then connected with clinicians for appropriate treatment or clinical trials. She stressed that the goal is to make these advancements accessible to everyone, not just those at tertiary medical centers.

The discussion also touched upon the importance of whole genome sequencing being integrated earlier in the diagnostic journey, noting that the cost has decreased significantly, making it more accessible. The potential for AI to connect seemingly unrelated aspects of a patient's health and reveal hidden diagnostic patterns was also discussed as a significant future development.

Making Diagnoses Accessible

The Manton Center's accessibility for self-referral was highlighted, encouraging families struggling with undiagnosed diseases to reach out. The key benefit of these AI tools, as emphasized by the speakers, is the reduction in time to diagnosis, which can drastically improve outcomes for patients and their families.

Call to Action and Future Outlook

The conversation concluded with a look towards the future, with plans to make the AI tools more accessible to a broader audience. The research, supported by the OpenAI Foundation, signifies a promising step forward in leveraging AI to alleviate suffering and bring clarity to those facing rare and undiagnosed medical conditions.