The Data Pulse

Rare diseases present a unique challenge in both diagnosis and treatment, given the small number of cases, often leaving them undiagnosed. I talk with Professor Isaac Kohane of Harvard Medical School about the Undiagnosed Diseases Network's efforts to catalogue and diagnose rare diseases, focusing on the role that data science--specifically, merging genotype and phenotype information, can play in bringing about hope to the families of affected individuals. We also discuss efforts to aggregate EHR data from hospitals across the world in order to better predict COVID-related symptoms and outcomes, faster than most healthcare systems have been able to.

Autism Spectrum Disorder remains a pressing yet elusive spectrum of conditions. In my conversation with Dennis Wall of Stanford University, we discuss the promise of technology- and augmented reality-based systems in both diagnosis and behavioral treatment for affected individuals. He shares the importance of understanding the context in which data is collected, as well as the ability for simple algorithms to yield actionable insights in the clinic.

In my conversation with Greg Ryslik, previously Chief Data Officer and now Special Adviser to Celsius Therapeutics, we talk about single cell sequencing technologies and the nuance they enable with respect to identifying and targeting the cell populations that are responsible for driving diseases, as well as the machine learning frameworks he employs when approaching problems with large-scale datasets.

Working with humans poses significant challenges to acquiring robust and complete data, but also remarkable opportunity, as I learn in today's episode with Professor Manish Desai of Stanford University. We discuss inferring causality from longitudinal data, clinical trial and observational study considerations, and the intersection of statistics and medicine at large.

Natural language processing has long yielded exciting predictions from word-based knowledge. Through my conversation with Zainab Doctor, Translational Science Head of nference, I chat about how we can now use text as a lens into the biomedical world, as well as the ability to integrate diverse data types to synthesize knowledge of all scales and yield clinically actionable insights.

Can we create new biological therapies with machine-guided design? Today I chat with Peyton Greenside, Co-Founder and CSO of BigHat Biosciences, on using machine learning to design therapeutic proteins, the advantages of using "smart" data over "big" data, and the importance of interpretability.

Today I discuss with Vyas Ramanan, of Third Rock Ventures and Maze Therapeutics, the evolution of the field's understanding of genetic modifiers and the role they play in disease, how drug discovery works when attempting to recapitulate protective effects of certain mutations, and how the convergence of certain advances in statistics and genomic sequencing and perturbation tools enable pressure testing of hypotheses at an unprecedented scale.

How should one think about building "the new wave" of biopharma teams? Turns out we can rely on principles from existing domains. In this episode, I talk with Ankit Gupta, CTO and Co-Founder of Reverie Labs, on systematization as a foundation, and on optimizing properties of a drug using machine learning and software-oriented frameworks.

Prescription digital therapeutics are emerging as an entirely new therapeutic modality. Corey McCann, President and CEO of Pear Therapeutics, chats with me about what they are, how they evolve as more data are collected, and the subsequent changes that are made to the treatment paradigms of neurological and psychological disorders. We also discuss what it took to get FDA approval for "software as a therapy," what skills are essential to succeeding in the space, and the growing presence of remote and asynchronous care.

What does an individual's digital presence reveal about imminent infectious disease outbreaks, obesity prevalence, and the spread of medical misinformation? How can Google Maps images reveal socioeconomic factors that contribute to disparities in health? When and how does community context matter? Elaine Nsoesie, Professor at Boston University School of Public Health, shares how she has led data-driven efforts to address these questions in the context of various diseases and communities around the world.

Molecular diagnostics are emerging as a precise way to detect diseases early on in prognosis. In today's episode, I chat with Harvard+MIT PhD Student Ava Soleimany on the role of data science in activity-based molecular diagnostics for early cancer detection and how the confluence of classification techniques with feature representation of cancer biomarkers can enable non-invasive, early detection.

Climate change has been noted as perhaps the greatest public health crisis of our times--Professor Francesca Dominici of Harvard University agrees. We discuss her data-driven efforts to demonstrate the negative health impacts from air pollutants, particularly for resource-poor communities, and how the current pandemic has further elucidated these disparities. She shares the steps necessary to move from academic science to changing policy, as well as her work in bridging the gender representation gap in STEM.

Deep learning has shown tremendous advances in making predictions from imaging data. Today I talk with Andy Beck, Co-Founder and CEO of PathAI, about deep learning specifically applied to pathology to both diagnose and lead to treatment options for patients, as well as the global implications that deploying such a technology could have.

In today's conversation, Gaurav Singal, who is the outgoing Chief Data Officer of Foundation Medicine, discusses the advent of molecularly-driven oncology in changing the paradigm of cancer diagnosis and treatment through precision medicine, the importance of using observational data, and the unpredictable nature of one's career that may yield a convergence of what were once disparate fields.

Gene therapies have emerged as a promising new modality for curing genetically-defined diseases; however, the naturally occuring variation remains limited. I chat with Sam Sinai and Jeff Gerold, Co-Founder/Lead ML Scientist and Head of Data Science, respectively, of Dyno Therapeutics about the role machine learning can play in better identifying and designing gene therapy vectors for a suite of traditional bottlenecks in the gene therapy development workflow, as well as the importance of building an environment where cross-talk between different domains is seamless.

Humans tend towards bias, but are our algorithms objective? Today I discuss fairness and bias in machine learning for healthcare with Professor Maryzeh Ghassemi of the University of Toronto. We delve into the ways in which bias pops up in the data that are used to train computational models, the particular dangers of systemic inequalities in healthcare being perpetuated by algorithms, and some of the steps needed to combat these deeply rooted issues.

Today I speak with Vineeta Agarwala, General Partner at Andreessen Horowitz and a physician at Stanford. She shares the importance of capturing time-course, evolving data on patients that range from the molecular to the clinical levels ("deep clinical annotation"), as well as key ingredients of a successful tech-biotech hybrid startup.

Conventional drug discovery ignores the spatial relationships between cells when assessing the effects of drugs. I talk with Imran Haque, VP of Data Science at Recursion Pharmaceuticals, about their data-first approach that combines "cell painting" with deep learning to assess changes in cellular morphology, thus providing a crucial lens into the molecular wirings of a given drug's mechanism of action. We discuss his team's work in repurposing therapies to tackle COVID-19 and what it takes to successfully merge deep learning with cellular images in a way that yields therapies.

Genomics has been a largely homogeneous field--both with respect to the researchers and the individuals whose data has been collected. Carlos Bustamante, Professor at Stanford University and F-Prime Venture Partner, whose work studies populations of diverse ancestry, claims that COVID has brought these disparities to light and that we have a unique opportunity to sequence the next generation of babies as one step towards more equitable precision medicine and drug development. He also discusses how the patient, as the healthcare consumer, will soon govern what solutions are adopted, including digital medicine and continuous tracking.

In 2016, a team at Google Health led by Product Manager Lily Peng accurately predicted diabetic retinopathy in patients solely from images of their eyes. Today I chat with Lily about this work in diagnosing diabetic retinopathy by tapping into advances in deep learning, what factors determine whether a medical problem is well-suited to machine learning tasks, and the iterative process to achieve model performance in the context of a clinical environment. We also discuss tactics to effectively deploy such technologies in resource-poor regions in a sustainable way.

What does a hybrid team working at the interface of machine learning and biomedicine look like? In this episode, I chat with Daphne Koller, Founder and CEO of insitro, about a data-first approach to drug discovery, building the systems that enable large-scale learning, and the importance of a bilingual culture in "digital biology".

Welcome to The Data Pulse, a podcast that explores the growing role that data science and computation play in the latest biotechnology and biomedical innovations. Join me as I speak with pioneers in the space, to learn more about how they use data-driven tactics to advance biology and medicine, to understand the principles that guide their work, and to better equip yourself with the language and thinking essential to effectively contributing to this space.

I’m your host, Anika Gupta, a PhD student in Bioinformatics at Harvard and the Broad Institute. To introduce the podcast, I'm joined by Anthony Philippakis, Chief Data Officer of the Broad Institute and a Venture Partner at GV (formerly Google Ventures). We discuss his unique journey and roles, why this intersection is promising, and what the key ingredients for innovation in this space are.