Moving Science into the ClinicSeptember 10, 2019
How Mount Sinai Drives Real-World Applications of Research Discoveries
By Douglas McCormick
From the fall issue of Mount Sinai Science & Medicine magazine
The long-anticipated combination of artificial intelligence and molecular biology is now producing personalized, precision treatments for patients at Mount Sinai, and Mount Sinai is more active than ever in helping to bring the benefits to patients worldwide.
Artificial intelligence (Al) and RNA sequencing at the Icahn Institute for Data Science and Genomic Technology at Mount Sinai have been used to test personalized treatments for multiple myeloma patients in a small-scale trial. Sema4, a health intelligence company spun out of Mount Sinai, uses Al-based algorithms to derive powerful insights that drive personalized clinical care solutions, most recently embarking on a study in collaboration with Mount Sinai to understand and predict risk factors for preeclampsia. In the Lillian and Henry M. Stratton-Hans Popper Department of Pathology, Molecular and Cell-Based Medicine at the Icahn School of Medicine, pathologists are training computers to read tissue slides and grade prostate tumor severity. At RenalytixAI, another company that utilizes Mount Sinai technology, artificial intelligence is being used to integrate blood-based biomarkers with features from a patient’s electronic health record (EHR) to create highly predictive clinical in vitro diagnostics for chronic kidney disease and kidney transplant rejection. The first AI-IVD test, KidneylntelX, received Breakthrough Device Designation from the FDA in May and is scheduled to become available at Mount Sinai in late 2019.
Mount Sinai has joined machine learning with advanced genomics, creating mechanisms to comb through terabytes of genomic sequences, clinical test data pathology slides, and clinical records to find new diagnostics and therapies, and then speed those patientcentered methods into the clinic.
Bringing Patient-Centered Medicine to Market
Mount Sinai Innovation Partners (MSIP) is the innovation and commercialization engine of the Mount Sinai Health System, advancing Mount Sinai discoveries to patients on a global scale. “Our group is the product of a strategic decision by the institution to focus on commercially relevant translational research that can bring breakthrough technologies, products, and services to everyday medical practice,” says Erik Lium, PhD, Executive Vice President, who has led MSIP since 2014. “The historical notion of ‘tech transfer’ is outdated, and it’s not what we do,” says Dr. Lium. MSIP is part venture investor, underwriting development of commercially relevant intellectual property, and part commercialization coach, providing guidance to help advance early-stage concepts or discoveries and translating these into products and services that have a positive impact on health care. MSIP’s 40-person staff provides experience in business development, new ventures, intellectual property, contracts and licensing, alliance management, finance and operations, marketing, and administration, plus the insights of”executives in residence” and other seasoned business advisors who can expand support for the development of translational assets into market-ready technologies. In addition to counseling and mentoring, MSIP manages several mechanisms for financial support:
- 4D pilot funding of $25,000-$40,000 supports specific experiments designed to generate early-stage proof-of-concept data for a program that could have commercial potential.
- Funding through the i3 Accelerator, a $10.5 million fund seeded by Mount Sinai, supports the development of commercially relevant intellectual property that will substantially de-risk a research program and position it for commercialization.
- In special cases, Mount Sinai may decide to make larger investments in programs exhibiting substantial commercial promise.
Teaching Computers To Read Slides
One of the newest medical AI projects emerging from Mount Sinai can be found in the Department of Pathology. Carlos Cordon-Cardo, MD, PhD, Irene Heinz Given and John LaPorte Given Professor and Chair of the Lillian and Henry M. Stratton Hans Popper Department of Pathology, Molecular and Cell-Based Medicine. and System Chair, Department of Pathology, and his collaborators set out in the mid-1990s to develop innovative tools to mathematically define and quantitate normal and disease structures seen in histology slides. Thanks to advances in molecular biology and computer science, success is near, he says. “Disease could be defined as the altered molecular expression of the landscape. Applied to pathology, this landscape or topography embodies the multidimensional analysis of tissues and cells by innovative techniques that provide objective and quantitative data for distinct features in diseased tissues compared to the corresponding normal elements. The result is greater precision and specificity of disease diagnosis in individual patients and predictability of response to alternative treatments,” says Dr. Cordon-Cardo.
Reading tissue slides requires highly specialized image-analysis algorithms that can be trained by experts to reproduce – and improve on – the sort of pattern recognition that pathologists do every day. The team’s first module implements the Gleason Score, the five-point scale for assessing prostate cancer. Changes in cell size, gland variability, and tissue architectural distribution fall into patterns that, with training, can be recognized and categorized by an experienced examiner.
But people are prone to fatigue and inconsistent: the same pathologist can sometimes give the same slide different scores on different days. Machines can be faster, tireless. and more consistent.
Dr. Cordon-Cardo and his collaborators developed mathematical methods to analyze histology-slide images and recognize important features. Once the computer has been taught how to “see,” it can be trained by human experts, who review the system’s judgments and correct mistakes. After a myriad of iterations, with thousands of slides’ worth of training and retraining, the system can reach more than go percent accuracy. Pattern-recognition subroutines developed for one cancer can be modified and retrained to work on new tumors, making the technologies a sort of Swiss Army knife for Al-driven pathology.
Patient Information Provides Road Map
While Dr. Cordon-Cardo and his team look for patterns in visual images, those at Sema4 look for patterns in diverse data. Eric Schadt, PhD, came to Mount Sinai in late 2011 as the Founding Director of the Icahn Institute for Genomics and Multiscale Biology and Professor and Chair of Genetics and Genomic Sciences. He arrived from Sage Bionetworks, the Seattle nonprofit he co-founded to focus on systems biology and data-driven research, and promptly founded Sema4 as an internal platform for sharing biological data and computing tools. driving medical discovery, developing diagnostics, modeling disease processes, and identifying treatment targets. In June 2017, Sema4 was spun off as a separate company, wholly owned by Mount Sinai with Dr. Schadt serving as the company’s CEO. Dr. Schadt concurrently serves as Dean for Precision Medicine and Mount Sinai Professor in Predictive Health and Computational Biology at the Icahn School of Medicine.
The company emphasizes multiscale biotechnology drawing together information on patients’ genomes, proteomes, proteins in the blood or tissues; transcriptome, from RNA messages to the protein-making machinery; epigenome, chemical “bookmakrs” on the DNA; and microbiome, the bacteria, fungi, and viruses that inhabit us all. Linking this information to EHRs pulls in valuable information from conventional diagnostic tests and physicians’ clinical observations.
The company’s network modeling capabilities are powered by Centrellis™, Sema4’s proprietary health intelligence platform. Centrellis also facilitates the delivery of novel test content and interpretations for state-of-the-art clinical testing, enabling more informed decision-making for physicians and patients. Sema4’s current products include inherited disease carrier screens for prospective parents, non-invasive prenatal testing, home cheek-swab tests for childhood conditions and drug reactions, and solid-tumor sequencing panels that assess 161 gene variants that can affect prognosis and treatment. More are on the way.
A Close and Personal Look
Inside the Icahn School of Medicine, the 500 staff of the Icahn Institute for Data Science and Genomic Technology also look for patterns in data, working on a broad range of projects in genomics, data analysis. oncology, immunology, cardiovascular medicine, and more. Their goal is to integrate machine intelligence, advanced biotechnology, and clinical expertise to discover and guide new kinds of personalized diagnostics and treatments.
Adam Margolin, PhD, who came to Mount Sinai in April 2018, is Director of the newly formed institute, Jean C. and James W. Crystal Professor and Chair of the Department of Genetics and Genomic Sciences, and Senior Associate Dean for Precision Medicine. He arrived with a focus on artificial intelligence and genomics after working at Oregon Health & Science University and Sage Bionetworks in Seattle, and with the aim of increasing the Department’s mathematical and data science capabilities.
The Icahn Institute’s largest current project, running in conjunction with Sema4, adds RNA sequencing to genomic DNA data to illuminate tumor cells’ idiosyncratic chemistry-and reveal targets for targeted chemotherapy. The approach is yielding results in treating multiple myeloma, an often-refractory malignancy of certain white blood cells. A recent small-scale trial applied machine-learning techniques to select optimal treatment regimens for 30 patients; of these, 25 saw clinical benefits.
Also rewarding, says Dr. Margolin, is the chance to discover new biological markers that offer both novel diagnostic opportunities and new ways of thinking about therapeutics. For example, a new machine learning technique, kernelized Bayesian transfer learning (KBTL), “allows us to solve multiple problems simultaneously and find common mechanisms or factors that drive tumors across tumor types.” When, for example, KBTL looked for multiple mutation signatures of 123 different genes in 10 different kinds of tumors, it found errors in one gene, called Fbw7, that showed up in tumor after tumor. The mutations appear to disable some of the cell’s usual energy production pathway, substituting new ones – and providing unique targets for disrupting the aberrant metabolism and killing cancerous cells in a variety of malignancies.
RenalytixAI: Transforming the Diagnosis and Management of Kidney Disease
A new idea for improving kidney diagnostics germinated soon after Steven Coca, MD, Associate Professor, Department of Nephrology and Associate Chair for Clinical and Translational Research, and Girish Nadkarni, MD, Assistant Professor, Department of Nephrology and Clinical Director of the Charles Bronfman Institute for Personalized Medicine, met at Mount Sinai in 2015. They shared an interest in kidney disorders, along with expertise in predictive blood and urine biomarkers, the untapped valuable information of EHR systems, and using machine-learning algorithms to analyze large-scale, disparate datasets. They soon focused on chronic kidney disease (CKD) and the lack of innovation in identifying patients with accelerating kidney disease earlier, to break the march to end-stage renal failure and dialysis.
“In treating CKD,” Dr. Coca says, “you’re working on a timescale of months to years,” enough time for clinicians to apply Al-supported treatments in real time.” The clinical management and health economics of CKD were such significant problems that we felt the time was right to introduce advanced diagnostic technologies like machine learning to start to change the paradigm.”
Says Dr. Nadkarni, “Today there are 30 million Americans with existing kidney disease, and most don’t know they have it or are not identified as fast progressors. Over 50 percent of patients with fast-progressing disease end up crash-starting on dialysis in the ER, never having seen a specialist. This is simply unacceptable.”
The pair researched machine-learning algorithms that could combine predictive blood proteins with key features in a patient’s EHR from some of the 44,000+ patient samples stored in Mount Sinai’s BioMe™ BioBank.
The results nearly doubled the positive predictive value for Rapid Kidney Function Decline, or RKFO, in patients with type 2 diabetes over the best standard-of-care diagnostics available today. During the research process, Dr. Nadkarni and Dr. Coca were introduced to diagnostic industry veteran James McCullough, who in turn was introduced to Mount Sinai Innovation Partners and Barbara Murphy, MD, Murray M. Rosenberg Professor of Medicine, System Chair of the Department of Medicine for the Mount Sinai Health System, and Dean for Clinical Integration and Population Health.
Dr. Nadkarni came up with the name “RenalytixAI” and in November 2017, a company was born to translate Mount Sinai research into a commercial, quality-controlled development process for an artificial-intelligence-enabled in vitro diagnostic, KidneylntelX. RenalytixAI licensed this key technology from Mount Sinai in June 2018 and completed a public offering on the London Stock Exchange in October 2018, raising $29 million in equity.
Development has been rapid, and the FDA is now in the process of evaluating KidneylntelX and its ability to identify RKFD as a primary end-point under Breakthrough Device Designation. RenalytixAI’s next innovation, consisting of a portfolio of advanced diagnostic and prognostic solutions in kidney transplant, grew out of work by Dr. Murphy. A leader in kidney transplant immunology, she applies a systems biology approach to improve matches between transplant recipients and donated organs, guide immunosuppression treatment, and predict postoperative complications. She and colleagues have identified multiple gene profiles that outperform conventional methods for predicting transplant rejection and long-term transplant survival. “Individual transplant recipients inherently respond differently toward a transplanted organ, yet we treat most patients in a similar manner,” Dr. Murphy notes.
With detailed RNA analysis, she says, transplant physicians can potentially stratify patients at baseline and monitor them post-transplantation, alerting the transplant team to the risk for an underlying acute rejection that may benefit from a change in the immunosuppressive drug regimen.