Through the use of Genomenon’s AI-driven Mastermind Genomic Search Engine, the Rare Genomics Institute succeeded were a leading genetics laboratory had failed in securing diagnosis for an ultra-rare disease patient. Romina Ortiz, Chief Operating Officer and Vice President, Patient Advocacy at the Institute explains how the platform could prove transformative in tackling the challenges of ultra-rare disease diagnosis.
Rare genetic diseases pose a challenge for diagnosis in three ways. First, there are fewer patients from which to collect data, so there is less known about them as compared with more commonly studied diseases. Second, clinicians are less familiar with the rare disease symptomatology and presentation, so their diagnosis may be delayed. Perhaps the most difficult challenge in diagnosis is that there tends to be less of a focus on the study of rare disease because there are less financial rewards than more common diseases, so the information about each rare disease is not as thorough or readily available.
Irrespective of the reasons, when it comes to diagnosis, it is critically important to ensure that whatever information exists about a rare disease, it is identified and incorporated into the patient’s diagnosis. For patients who undergo genomic profiling, the industry-accepted guidelines for interpreting the clinical significance of genetic variants require that the following information be considered in addition to the clinical scenario of the patient:
Due to the vast amount of published information, finding the scientific publications – sometimes a single article – containing the information needed to unlock the diagnosis for a patient with a rare disease is very challenging.
How does the Mastermind Genomic Search Engine work?
Mastermind is an index of the genomic literature – all the publications that contain information about genes or genetic disease, including rare genetic disease. Mastermind indexes millions of published scientific articles, looking for mentions of thousands of different human diseases, tens of thousands of human genes, and literally billions of possible genetic variants, no matter how an author may describe them in their text, figures, or tables. Mastermind does this automatically with its proprietary Genomics Language Processing (GLP) capability – an artificial intelligence (AI) text analysis technique focused on the complexities and idiosyncrasies of genetics and pathology.
This index of genomic literature is organized for easy retrieval by Mastermind users to query if a potential causal variant has ever been seen before, and if so then in what context and citation. The querying and resulting information are presented with the end-user in mind which goes beyond doing Google or PubMed searches.
The variant analysis and interpretation process described above tends to be time-consuming and error-prone. Tools like Mastermind can be critical in finding the causal variant and associated information which is key in diagnosis.
Using the standard search tools, the most common of which are PubMed or Google Scholar, scientists run a risk of missing critical citations that may contain the disease-gene-variant information necessary to accurately diagnose their patient or would otherwise have to spend many hours scouring the search results looking for the right information.
Mastermind solves these challenges by providing a knowledge base comprised of curated and comprehensive genomic articles with live updates to capture literature being published. The organization of the information in the knowledge base is powerful. Often in scientific research, the tools which bring efficiency in time and indirect cost in labor are the ones which add value.
As an example, recently scientists at Rare Genomics (RG) patient advocacy group undertook a similar research project where they had to re-analyze a patient’s whole exome which had been previously analyzed by a leading genetics laboratory as part of the patient’s long diagnostic odyssey. The lab was unable to find any clinically relevant genetic mutations that could provide a diagnosis. The re-analysis of this patient’s whole exome by RG scientists included the use of Mastermind search engine and the research revealed a single published report in the scientific literature that matched the patient’s DNA data. The patient in the cited literature shared similar symptoms with the patient being analyzed. With this finding, RG recommended that the patient be re-examined based on the diagnosis found in scientific research.
It’s not just beneficial for patients
It is not often discussed – at least not in the same breath as we discuss the accuracy and acuity of patient diagnosis – but healthcare is a business. As such, the business of healthcare has strong financial considerations that drive decision-making.
While there is great potential in using next-generation DNA sequencing (NGS) to diagnosis more and more patients and patient types, the challenge of finding enough highly-skilled, expert variant scientists to curate the evidence needed to interpret their patients genetic data is still the rate-limiting factor in fully scaling the use of genomics for precision medicine.
As whole exome or whole genome-based testing becomes increasingly economically feasible to understand and diagnose rare diseases, studies have been published to show the direct impact of cost savings inpatient care. Technologies which facilitate the generation of the genomic data and subsequent analysis and interpretation are directly linked to these cost savings. In a recent study (Farnaes, L et al, Nature Genomic Medicine 3, article number 10, 2018) published by the Rady’s Children’s Hospital in San Diego, a cohort of acutely sick babies who received genomic testing and diagnosis helped save hundreds of thousands of dollars in healthcare. Moreover, beyond cost savings, the likelihood of death was reduced by 43% in the same cohort of neonates – and that is immeasurable. Rady’s Children’s Hospital also uses Mastermind for their genomic research.
Modern data mining technologies underpinned by machine learning and/or AI facilitate quick decision making while improving the accuracy of results. We strongly believe that whereas AI will never truly replace human review, the advanced and informed AI algorithms that Genomenon employs in building the Mastermind database will continuously reduce the manual process of finding, organising, and annotating the evidence needed to interpret patients’ genome data to the point where truly scalable genomic medicine will be a reality.
Tackling tomorrow’s hurdles
Medicine is still a conservative discipline; when a doctor’s name is on the line in a patient’s clinical report that will drive critical healthcare decisions for the patient, there is every reason to be precise. The main challenge is not the application of AI, it lies in demystifying and unpacking the term for health practitioners so they are able to understand and accept that these technologies are complementary tools.
A majority of the data analysis and decision support tools we see today remain for research use only. We still rely on human expertise and knowledge to make the final decision when it comes to patient care. However, what we are witnessing today in healthcare is the increasing acceptance and use of AI-based technologies that enable human decision-making to become faster and more accurate. This is particularly true when decision-making is dependent on having an understanding of the complete landscape of literature published for a specific disease, and that too in the correct context. This is where we see applications of machine learning and AI to complement and facilitate human decision-making that can be rationalized and explained.
Specific to the Mastermind application, the software company addresses these challenges by:
Mastermind is not just providing users with data one variant at a time for a single patient. It has a vast repository of genetic and genomic information, including how genes and variants play a role in disease. Genomenon uses this data to help pharmaceutical customers better inform their drug discovery process with a landscape view of the pathogenic mechanisms of disease. Pharmaceutical and biotech companies leverage this information to improve the success of their candidate selection process and clinical trial enrollment procedures, as well as to improve their ability to repurpose their drugs for additional indications.