Benjamin Glicksberg, PhD, is an Assistant Professor within the Department of Genetics and Genomic Sciences and a member of the Hasso Plattner Institute for Digital Health at Mount Sinai. Dr. Glicksberg joined Mount Sinai in October 2019 as the Institute’s first faculty recruit. His research activity lands at the intersection of precision medicine, bioinformatics, data science, and machine learning. Dr. Glicksberg’s work leverages multi-modal data such as genomics, clinical data drawn from electronic health records (EHR), and tech devices, with the ultimate goal of optimizing health. Dr. Glicksberg joined Mount Sinai from the University of California, San Francisco (UCSF), where he was a postdoctoral scholar within the Bakar Computational Health Sciences Institute.
At UCSF, Dr. Glicksberg led analysis on a real-world evidence project sponsored by the U.S. Food and Drug Administration, using retrospective, real-world data as support for a device label expansion study. In this study, Dr. Glicksberg examined data from retrospective hospital records and determined, based on a carefully controlled design, whether a medical device used in adults with a rare disorder related to chest deformities could also be approved for use in pediatric populations. This s tudy allowed researchers to recommend label expansion of the device without needing to work within the typical randomized control trial framework, serving as a useful alternative to expedite device access to a small and protected patient population.
In his postdoctoral work, Dr. Glicksberg built two software applications that improve replicability and lower barrier of entry to EHR research. Enhancing the commonly used Observational Medical Outcomes Partnership (OMOP) data model, he built ROMOP, an open-source R package that allows researchers to easily connect, search for patients, and pull cohorts of data in a machine-learning-ready format. While ROMOP was built for those with some expertise in coding and working with EHR data, Dr. Glicksberg also built an application for researchers with no prior coding or programming expertise. The application, PatientExploreR, is a visualization tool that allows for interactive “point-and click” interfacing with OMOP-formatted EHR data.
At Mount Sinai, Dr. Glicksberg is interested in collaborating with the divisions of cardiology and psychiatry/neurology to provide his expertise in machine-learning algorithms to make each clinical domain more data-driven. Mount Sinai has one of the best cardiology departments in the country, treating a high volume of diverse patients. As such, the large amount of quantitative clinical data generated is rife for machine-learning initiatives, especially routine cardiac procedures (EKG/ECG) and echo imaging. These data points, coupled with EHR data and genomics, may identify signals that can be used to personalize care. As for psychiatry/neurology, many phenotypes in mental illness and psychiatry prove harder for machine learning, as these phenotypes do not have clear, quantitatively described biomarkers; the majority are qualitative, and rely on imperfect representations such as clinical impressions, questionnaires, or scales, which do not necessarily reflect underlying biology. By introducing machine-learning algorithms, Dr. Glicksberg hopes to further quantify this field, with the goal of improving data-driven science behind important clinical insights such as treatment selection, why certain interventions do and do not work, and understanding readmissions.
The Glicksberg Lab focuses on a multitude of research areas that seek to transform how we receive care.
Countless combinations of therapeutic options exist for the vast expanse of human disease. Identifying the right medication (or medications), at the right dose, at the right time is challenging. Incorporation of each individual’s unique comorbidities, circumstances, genetic makeup, and cost-benefit probability is even more difficult. We identify clinical scenarios which data-driven statistical and machine learning approaches may identify complex patterns, helping clinical practitioners help their patients. These projects range from predicting readmissions to surgical outcomes.
The traditional drug discovery process is extremely time consuming and expensive. The field of drug repurposing attempts to find novel indications for therapeutics which are already in use and FDA-approved for other indications, thereby facilitating faster and cheaper implementation of drugs into practice. Computational pipelines have been shown to significantly augment the drug repurposing process. Here, large-scale drug and disease related information, including genomics, transcriptomics, proteomics, and chemical features are compared to determine data-driven novel connections. For example, drug and diseases signatures are created from specific differential gene expression profiles and compared against one other with the goal of identifying drug profiles that reverse the signature of the disease. We perform studies in a variety of clinical domains to try to identify novel treatments for diseases via this strategy and beyond.
While complex diseases, like type 2 diabetes mellitus, are often considered a single entity, the wide variety of heterogeneous manifestations and underlying causes suggest that there might be underlying substructures which may benefit from more precise treatment. By coupling multiple modalities of patient data of those suffering with complex diseases and using state-of-the-art machine learning techniques, disentangle any patterns that exist in underlying etiology and/or groups of clinical manifestations. To this end, we use topological, dimension-reducing techniques like clustering and variational autoencoders.
Using computational pipelines to analyze heterogenous biomedical big data has produced countless impactful findings. Yet lack of knowledge of various programming languages and domain-specific ontological nuances make it difficult for others in the scientific and medical community to utilize these pipelines. It is clear that creating web app, software, and packages to help facilitate those without technical and/or domain knowledge to perform such analyses can benefit many areas of biomedical research. We create software and tools to enable various aspects of computational biomedical-related workflows, such as exploring EHRs, perform drug repurposing experiments, or annotate radiographs.