BIDMC Pathology’s Schlesinger Lecturer Aviv Regev, PhD, Discussed “From Cell Atlases to Medicines, with AI”
Aviv Regev, PhD, Head and Executive Vice President of Genentech Research and Early Development, presented “From Cell Atlases to Medicines, with AI” on May 12, as the annual Monroe J. Schlesinger Lecturer, one of two named lectures sponsored by the BIDMC Department of Pathology each year. She leads an active research lab focused on developing and applying experimental methods and computational algorithms to decipher intra- and intercellular circuits within tissues. Dr. Regev earned her PhD in Computational Biology from Tel Aviv University. She is also a founding co-chair of the Human Cell Atlas.
Dr. Regev walked the audience through her more than 10-year journey to discover how AI could find patterns in the layers of data from cells, genes, tissues, and their connections. She explained that in order to treat disease we need to know how our genes are expressed in the body’s cells, and what happens when the genes are mutated or their expression changes in disease. While the idea is simple, the problem is scale. The number of hypothetical combinations of cells, genes, and their variants is astronomical, even more than atoms in the universe.
Previously, researchers’ approach was focused on quality–to take one biological sample and understand everything about it. She asked the audience to consider a different approach. That if you have a large enough quantity of data, even if each measurement is noisier, it can become a new kind of quality. Regev’s team uses a method called “lab in the loop”: they start with a large-scale experiment, where they measure, for example, the gene expression of a large quantity of cells, use the data to train an AI model, then ask the model to suggest the next experiment, and repeat. This approach builds on breakthroughs in single-cell genomic technology, which can capture information about millions of cells, rather than taking one cell and trying to learn everything about it.
AI models can help fill in the gaps in many different ways. Some examples Dr. Regev described included:
- Predicting gene expression based on genome sequence: This has been a challenging problem to crack since biologists started studying gene regulation many decades ago. To attack the problem at scale, Dr. Regev’s group measured the impact on expression of 100 million random DNA sequences. This experiment provided the data to train an AI model to predict the impact on expression of other natural and random DNA sequences.
- Understanding the role of genes in disease: They perturbed genes in primary immune cells and observed the results for millions of individual cells, and then connected this to human genetics data to help determine where disease starts.
- Predicting therapeutic interventions: They asked the AI model to predict which genes or, combination of genes, should be targeted to return a diseased cell to a healthy state.
AI models can help fill in the gaps in many different ways. Some examples Dr. Regev described included:
- Predicting gene expression based on genome sequence: This has been a challenging problem to crack since biologists started studying gene regulation many decades ago. To attack the problem at scale, Dr. Regev’s group measured the impact on expression of 100 million random DNA sequences. This experiment provided the data to train an AI model to predict the impact on expression of other natural and random DNA sequences.
- Understanding the role of genes in disease: They perturbed genes in primary immune cells and observed the results for millions of individual cells, and then connected this to human genetics data to help determine where disease starts.
- Predicting therapeutic interventions: They asked the AI model to predict which genes or, combination of genes, should be targeted to return a diseased cell to a healthy state.
As these approaches are applied across a broad range of cells and biological systems, and across multiple scales from genes to cells to tissues, they build toward the idea of a foundation model of cell and tissue biology: an AI model of a system that can make predictions — of gene sequences, experiments, or potential treatments — that have not yet been tested in the lab or clinic, but could accelerate the development of biological understanding or new therapeutic approaches. Already in her work, Dr. Regev and her colleagues are using this Lab-in-the-Loop approach to design new potential therapies including antibiotics, antibodies, and even cancer-treating vaccines. By uniting large-scale human biology data and AI, foundation models provide a new approach to developing medicines for patients.
The annual lecture is named for Monroe J. Schlesinger, MD who was a distinguished pathologist and the first Chair of the Department of Pathology at Beth Israel Hospital in 1928. Current Pathology Department Interim Chair Vikram Deshpande, MD, noted that Dr. Schlesinger was a pioneer in bacteriology and immunology, as well as the first to recognize the severe nature of coronary disease in patients with diabetes.
