Abstract
Held in early September in La Jolla, California, the fourth annual AGBT Precision Health meeting brought together experts in genomic medicine to present the latest breakthroughs and cutting-edge research in the field. Together, they covered genomics, AI, variant detection, data sharing, patient stories, and more. From Katherine High, M.D.'s perspective on gene therapy as the president of Spark Therapeutics to fireside chats with patients and everything in between (including a statistics lesson from Leslie Biesecker, M.D., senior investigator at the NHGRI), the conference took a multi-pronged approach that offered a deep dive into the world of precision medicine.
Women kick things off at AGBT
Unlike a music concert, when opening acts crescendo into the headliner, scientific conferences tend to eschew anticipation by opening big—right at the beginning. The AGBT precision health meeting was no exception. Not only were three of the most captivating speakers in the opening session—they all happened to be women.
The session began with Prof. Dame Sue Hill, Ph.D., chief scientific officer for the National Health Service in England, presenting on “The Power of Precision Health—Delivering Genomics in the NHS.” Hill detailed the ambitious work surrounding the national genomic infrastructure in England, the framework for personalization, the strategic focus over the next 10 years, and the plans to implement a genomic future. After a presentation chock full of lessons learned and future goals, Hill ended with a prophetic quote from Marie Curie, “Nothing in life is to be feared, it is only to be understood.”
Heidi Rehm, Ph.D., chief genomics officer, Center for Genomic Medicine and Department of Medicine and Medical Director of the Clinical Research Sequencing Platform at the Broad Institute, spoke next, giving a talk entitled, “Curating the Clinical Genome at a Global Scale.” Rehm started with a patient story (covered in detail in 2015 in The Atlantic) in which her clinical genomics lab detected a PTPN11 variant in a fetus with increased nuchal translucency on the ultrasound. The variant had been reported as pathogenic previously, as it had been seen in a patient with Noonan syndrome. But Rehm was denied access to a research dataset to determine the allele frequency. Although it was later determined the variant was benign, the couple had already terminated the pregnancy. Rehm asserts that data silos need to be broken down in order for genomic medicine to be effective. “For the best outcomes for patients, we need to share data on a global scale and critically evaluate evidence as a collective community,” Rehm noted.
Rehm described the resources currently available to researchers, such as ClinGen, a consortium of people sharing data, and the ClinVar database. ClinGen asks three critical questions: is this gene associated with a disease, is this variant causative, and is this information actionable? As of August, ClinGen had 880,033 submissions on 560,351 unique variants from 1,320 submitters in 67 counties. Another resource Rehm described is the Matchmaker Exchange, launched to address the challenges of connecting case data that sits in isolated databases to determine the genetic causes of rare diseases.
Rounding out the opening keynote session, Elaine Mardis, Ph.D., co-executive director of the institute for Genomic Medicine and Nationwide Foundation Endowed Chair in Genomic Medicine spoke about “Integrating Genomics into Precision Pediatric Cancer Medicine.” Mardis stressed that there are challenges in cancer pediatrics that are not present in the adult setting. Bringing precision medicine to pediatric cancer is one of her main goals in her current role as president of the American Association for Cancer Research (AACR). She added that the current therapies available to treat cancer can give children significantly poorer quality of life and elevated risk of more cancer as an adult which, she noted, is “unacceptable.” Because each child is essentially an “N of one,” she stressed that the need for data sharing is paramount.
Eric Topol M.D., director, Scripps Research Translation Institute
Data sharing with AI
Eric Topol, M.D., director of the Scripps Research Translational Institute continued on the theme of the importance of data management, starting his talk entitled “High Performance Medicine” by noting that “we have well-exceeded our ability as humans to deal with data.” He added, “we need help from machines” and to have high-performance medicine, he asserted, “we need high-performance computing.”
Topol noted that there is a problem with the term precision medicine, which is not very precise. Rather, we need medicine that is highly accurate and highly precise. But, in order to get there, we first “have to fess up about how inaccurate or imprecise our medicine is today.”
Pointing out some “interesting factoids” about the imprecision of medicine, Topol noted the large number of diagnostic and medical errors. The diagnostic accuracy if a physician thinks of a diagnosis in the first 5 min of a patient is 98% he noted, but that number drops down to 25% if the diagnosis doesn't occur within the first 5 minutes. Also, he explained that when a physician (premortem) said that they were completely certain of a diagnosis, the autopsy found that they were wrong 40% of the time. Topol noted that this is neither accurate nor precise.
Topol then asked the audience how many believed that polygenic risk scores (PRS) should be used in clinical practice? When only a few hands were raised, he said, “that's too bad.” Understanding that some in the community are “anti-PRS,” he made the case as to how PRS can help bring medicine deeper than where it is today. Topol mentioned an app developed by Ali Torkamani, Ph.D., associate professor and director of Genome Informatics at the Scripps Research Translational Institute, and others called MyGeneRank. The app, when given permission, can access genetic data from 23andMe to calculate a risk score for coronary artery disease (CAD). To date, there are more than 4,000 users. He said that using PRS in this way is “an inevitable part of how we can help discriminate the use of certain, actionable tests or therapies.”
Sophia Genetics combines data sharing and AI
“If you want to use Sophia, you have to share,” noted Kevin Puylaert, general manager North America, at Sophia Genetics. The goal at Sophia, one of the sponsors of the AGBT precision health meeting, is to “democratize data-driven medicine,” according to Puylaert. Sophia does not collect samples or perform sequencing. Rather, hospitals send their data to Sophia's AI platform for genomic variant detection, annotation, and interpretation. The report is then returned to the hospital to inform the clinician's diagnosis.
During a talk by Puylaert, he described how AI is at the heart of the platform, which uses statistical analysis, pattern recognition, and machine learning to analyze the data. Training its algorithm on the sample types and the sequencing and enrichment technologies, Sophia can analyze many more patients and collate the information in a manner that surpasses human ability.
Perhaps even more critical to the company's success, however, is the number of patients in its platform. “If you want to have a smarter AI, you need more data,” noted Puylaert. Sophia has a lot of data—working with 1,000 institutions in 81 countries and more than 380,000 genomic profiles, analyzing more than 15,000 cases a month.
Alexander Hoischen, Ph.D., associate professor in the department of human genetics and internal medicine at Radboud University Medical Center, Nijmegen, The Netherlands.
Whole genome sequencing is missing the “W”
On the second morning, Alexander Hoischen, Ph.D., associate professor in the department of human genetics and internal medicine at Radboud University Medical Center, Nijmegen, The Netherlands, presented a talk sponsored by Bionano Genomics on the importance of long read sequencing and long read optical mapping in genome analysis. He started by noting that structural variants are an underappreciated source of genetic variation.
Hoischen's talk, “Long-Read Sequencing and Long-Read Optical Mapping Complement Medical Genetic Assays to Reach Full Genome Analysis” focused on overcoming the limitations for studying structural variants using short read lengths commonly produced by NGS sequencing. The limitations may, he noted, “significantly contribute to the diagnostic gap in patients with genetic disorders who have undergone standard NGS, like exome or genome sequencing and cytogenetic assays like CNV-microarrays.”
Focused on intellectual disability (ID), Hoischen uses different methods to analyze 1,500 ID patients. Using different genomic methods including microarrays, exome sequencing, and WGS, Hoischen's group can diagnose 62% of ID patients. But, how can they reach the other 38%? What have they missed? He presented two possibilities: the mutation was not sequenced, or not detected. Based on those hypotheses, Hoischen focused on how long-reads, SMRT-sequencing and optical mapping techniques could fill the gap.
Hoischen showed that that SMRT sequencing detected ~25k SVs per genome and >30k indels per genome, the majority of which (60-70%) were not detected by standard NGS approaches. Long-read mapping using Bionano's Saphyr identified ~5500 SVs per genome (>500bp). They were able to, using optical mapping, allow access to the most difficult regions of the human genome. Because of their success in identifying variants, Hoischen wonders if these methods can replace cytogenetics entirely. In his view, “long-read sequencing harbors the potential to deliver near-perfect genome analysis in the future and long-read optical mapping may replace almost all karyotyping, FISH, and CNV-microarray assays in routine diagnostics.”