Publication date: 3 Nov 2021
Artificial intelligence (AI) will inevitably revolutionize the diagnosis and treatment of rare diseases. Taken as a whole, the number of rare diseases, their diverse characteristics and complex causes comprise an ever expanding data set that will require computational help to decipher. An estimated 40% of rare disorders are misdiagnosed at the outset, leading to incorrect management and treatment delays. As rare conditions are frequently first noticed in young children, the consequences of misdiagnosis and delayed treatment can profoundly impact disease severity over a lifetime.
What is AI, and can machines actually think? Computers can only mimic human intelligence by “learning” to recognize recurring patterns in large sets of data and make statistical classifications that can have predictive power. Error corrections in predictions can continually improve their accuracy. Being able to do this without human intervention seemingly appears intelligent. One increasingly common example is how an automated smart assistant on a mobile phone can learn to respond to your voice commands.
How can this power be harnessed to tackle the challenges posed by rare diseases? The sheer quantity of information on rare diseases is rapidly growing, and no single person can ever get to know and understand it all. Medical professionals already depend upon access to compilations of accurate, relevant and actionable electronic information to make diagnoses and therapeutic decisions. An increased utilization of AI merely refines and builds upon this concept.
The types of data that can be utilized to characterize rare diseases are diverse and complex. For diseases that have a genetic cause, entire genomic DNA sequences can be used to pinpoint disease-causing variants and diagnose new patients. Sifting through the genomes of millions of people may be necessary to establish a rare disease association. DNA sequencing alone may not be adequate for establishing a diagnosis, as some diseases are caused by non-genetic factors.
For other diseases, searching for patterns in the occurrence of a condition in certain geographic locations or links to specific environmental factors can unveil valuable clues. Information on symptoms, diagnostic test results and other observable markers can further define disease classifications. Even computer-based facial recognition technology is being recognized as a potential rapid diagnostic tool for some rare conditions. Assisted by AI, researchers can mine massive collections of data to uncover disease correlations and causes.
Expanding the observations to include those made in animals or simpler organisms can help unravel disrupted biological pathways that lead to disease development. Understanding such pathways is a key to developing new targeted therapeutics. The ability of AI to rapidly model and compare candidate drugs can accelerate their introduction to patients. Once a drug is in clinical use, there will be a need to monitor its efficacy and side effects, for which AI can play an indispensable role.
Work in all of these areas is currently underway, but huge challenges remain. First, the efforts are widely spread across international boundaries, academic institutions, governmental agencies and private companies with limited resources and little coordination. While this ensures a broad-based approach, progress will be fragmented. There may be no universal language, common terminology or criteria to describe many rare conditions. The quality of the source data may be inconsistent, leading to a “garbage in, garbage out scenario” with regard to utility. As mentioned above, the data itself can be DNA sequences, statistical correlations, diagnostic test results, images or stand alone empirical observations. AI may be necessary to sort and interpret such diverse data sets. Building better algorithms and infrastructure are essential to advancing the field.
Assuming that rare disease databases can and will be consolidated in the future, who will own and control such resources? Will it be governmental agencies, private companies or will it exist freely available on the web? What rights do patients have over their own data that may become part of a larger permanent database? Can they be compensated? How can their identity and privacy be protected? Much like the ability of criminologists to identify relatives of murder suspects from a forensic DNA sample, can a rare disease database find susceptible distant cousins without their knowledge? The answers to such questions are yet to be determined.
Under ideal circumstances, where may the marriage of AI with rare disease databases lead? Optimistically, it can make all known rare diseases diagnosable with some established guidelines for management and treatment. It can shorten the time to diagnosis, opening the door to interventions before a condition becomes severe or fatal. Finally, it may hasten a time when a physician anywhere in the world can readily access information to effectively diagnose and treat a rare disease patient efficiently with no barriers. This would be a tremendous leap towards a futuristic vision of precision medicine for all patients.