ai is ready to take on a massive healthcare challenge
The Significant Economic Impact of Rare Diseases
Determining the disease with the greatest annual economic impact might lead one to consider conditions like diabetes, cancer, heart disease, or obesity. However, the correct answer is actually rare diseases.
In 2019 alone, the financial strain imposed by rare diseases reached a substantial $966 billion. This figure significantly surpasses the costs associated with diabetes ($327 billion), cancer ($174 billion), and heart disease ($214 billion), as well as other chronic illnesses.
Understanding the Scope of Rare Diseases
The reason rare diseases are often overlooked is their individual low prevalence. A disease is classified as rare if it impacts fewer than 200,000 individuals.
Despite this, the cumulative effect is immense. Thousands of distinct rare diseases exist, collectively affecting approximately 400 million people globally.
Challenges Faced by Patients
A disproportionate number of individuals affected by rare diseases are children, representing roughly half of all patients.
The journey to diagnosis is often protracted and arduous. Patients typically endure a diagnostic odyssey spanning five years or longer, involving numerous medical tests and consultations with multiple specialists before a definitive diagnosis is achieved.
This lengthy process highlights the difficulties in identifying and treating these rare conditions.
The Diagnostic Journey is Becoming More Accessible
Previously considered an extremely difficult undertaking, significantly decreasing the length of the diagnostic process and its related expenses is now achievable. Approximately 80% of rare diseases have a genetic basis, and the convergence of technological progress and artificial intelligence is broadening access to genetic testing.
Whole-genome sequencing, a sophisticated genetic examination of complete human DNA, currently has a cost of less than $1,000. Illumina, the leading company in this field, is aiming to reduce the cost to $100 per genome in the coming years.
However, a substantial hurdle remains: the interpretation of this data within the framework of human health, a complex task. A typical individual possesses 5 million distinct genetic variations, and pinpointing the single variant responsible for a disease requires careful analysis.
Fortunately, recent breakthroughs in cognitive AI enable the automated examination of an individual’s entire genome sequence. This allows for the identification of disease-causing factors, effectively enhancing the capabilities of medical professionals.
This technology is poised to transform how we approach rare disease diagnosis.
Key Benefits of Advancements
- Reduced diagnostic timelines.
- Lower healthcare costs.
- Increased accessibility to genetic testing.
- Enhanced accuracy in identifying disease-causing variants.
The ability to rapidly and accurately analyze genomic data represents a significant step forward in personalized medicine. It offers the potential to improve patient outcomes and alleviate the burden of rare diseases.
The Transition from Specialized to Cognitive Artificial Intelligence
Achieving a widely applicable AI solution necessitated a fundamental shift in approach, moving away from narrow machine learning models towards more comprehensive ones. Researchers analyzing genomic information must sift through a vast number of data points, originating from diverse sources and presented in varying formats.
Currently, a complete human genome analysis can require up to eight hours, and the number of qualified scientists globally remains limited to just a few thousand. With the anticipated reduction in genome sequencing costs to $100, projections suggest that 50 to 60 million individuals will have their DNA sequenced annually. How will this surge in data be analyzed within the context of individual health? This is where cognitive intelligence offers a solution.
Cognitive intelligence, encompassing cognitive computing solutions, integrates artificial intelligence technologies such as neural networks, machine learning, and natural language processing to emulate human-like intelligence. These advanced AI models possess the capability to make informed decisions in intricate and contextually rich scenarios.
The majority of artificial intelligence systems currently in use are narrow in scope, designed for the execution of a single, specific task. For example, Amazon employs a narrow AI model to assess the freshness of produce prior to shipment. This is a valuable application focused on enhancing customer satisfaction and minimizing waste.
However, applying narrow AI to the challenge of genomic data interpretation proves nearly impossible. The analytical process undertaken by human experts is inherently complex. It is heavily reliant on context and involves numerous variable input parameters. Furthermore, each patient case demands the compilation of scientific evidence, often found in recent publications and across numerous databases, adding a significant time burden that necessitates a machine learning-based solution.
The complexity is further amplified by the dispersed nature of clinical evidence. A particularly intricate case processed on the Emedgene platform involved a 19-year-old patient suffering from retinal degeneration. After an initial negative assessment by a clinical laboratory, the cognitive AI model pinpointed variations within the POC5 gene, a member of the centriolar protein gene family.
Given the association of two genes within this family – CEP290 and POC1B – with retinitis pigmentosa, a prevalent form of inherited retinal degeneration, the cognitive AI proposed POC5 as a potential candidate. It established connections between variants, gene and patient phenotypes, and other established genes, a process that would consume considerable time for a human analyst. Subsequent animal studies validated that mutations in POC5 lead to diminished visual motor response.
Training narrow artificial intelligence models to effectively address complex patient cases involving multi-step workflows presents a significant hurdle. A “brute force” approach, utilizing extremely large datasets, could be considered. However, patient data is frequently fragmented, and organizations attempting to train models would struggle to access sufficiently large datasets, even assuming enough patients have been sequenced.
Even when employing a supervised learning methodology, annotating scientific data at scale poses a considerable challenge, as the same biomedical professionals required for genomic data analysis would be best suited for the annotation process.
Furthermore, locating and organizing the necessary genomic data for interpretation introduces additional complexities, requiring a distinct set of machine learning technologies. To illustrate the difficulty, consider the term "M protein." This refers to Protein M, but also to the virulence factor produced by streptococcus, and sometimes to myeloma protein.
Moreover, MYOM2, the gene encoding Protein M, is also occasionally designated as Protein M. Disambiguating entities within biomedical literature is a challenging task. Coupled with the difficulty of identifying positive, negative, or other relationships between entities, and the need to adopt hierarchical ontologies of medical terms, this presents a substantial challenge in itself.
When confronted with the intricate challenge of scaling genomic interpretation, cognitive AI systems appear to be the optimal solution. A synergistic combination of machine learning algorithms capable of performing genomic analysis, locating and organizing scientific and genomic data, and providing evidence-based interpretation of patient genomic data can significantly reduce the manual effort involved, thereby enabling scalability in precision medicine initiatives.
Deloitte’s technology trends report categorizes AI technologies that predict, prescribe, augment, and automate as disruptive forces. They advocate for the utilization of cognitive technology to reveal insights and connections within disparate data, particularly in areas where human decision-making is not easily scalable. Fortune Business Insights forecasts the global cognitive computing market to reach $72.26 billion by 2027.
The Role of Machine Learning in Genomic Medicine
Unlike autonomous vehicles, the expectation isn't for machine learning models to perform patient diagnoses without human intervention in the foreseeable future.
Instead, these models are designed to assist overburdened precision medicine initiatives in managing an increasing number of patients.
Accelerating DNA Analysis
Cognitive computing can be leveraged to provide a precise initial assessment of human DNA within a clinical setting, always under the guidance of qualified professionals.
This approach substantially decreases the time required for analysis.
Decoupling Growth from Personnel
The implementation of these technologies effectively separates the expansion of genomic medicine from the necessity of continually increasing the number of biomedical specialists.
This allows for scalable growth without being limited by workforce constraints.
Human Oversight Remains Crucial
It’s important to emphasize that these systems are intended as assistive tools.
Human oversight is a fundamental component of the process, ensuring accuracy and responsible application of the technology.
Benefits of Cognitive Models
- Reduced analysis time for DNA sequencing.
- Improved scalability of precision medicine programs.
- Support for existing biomedical professionals.
These benefits contribute to a more efficient and accessible healthcare system.
The Expanding Applications of Advanced Diagnostics
The convergence of advanced technologies is not only accelerating the diagnostic process for individuals with rare diseases, but also presents opportunities to enhance our comprehension and therapeutic approaches for more prevalent conditions.
Historically, insights linking rare genetic variations to common illnesses have largely been accidental discoveries.
Consider, for instance, the finding that individuals carrying loss-of-function mutations in the PCSK9 gene demonstrate reduced levels of serum LDL cholesterol and improved cardiovascular outcomes.
This observation directly contributed to the creation of PCSK9 inhibitors, which have proven to be superior to statins in lowering cholesterol.
Extending Precision Medicine
The very technology enabling large-scale precision medicine for rare disease diagnosis can be adapted to pinpoint the genetic factors influencing desired responses to pharmaceutical interventions.
This represents a significant shift towards a more personalized and effective approach to treatment.
A Call to Action
Having successfully transformed rare disease diagnostics, the focus should now be on widespread implementation.
Recent advancements in both genomic sequencing and artificial intelligence are facilitating the integration of genomic medicine into mainstream healthcare systems.
Over 400 million individuals affected by rare diseases are currently awaiting diagnosis, and a substantial number of patients with common ailments stand to gain from these innovations.