Artificial Intelligence

Of the millions of genetic differences between people, only a small fraction contribute to disease—but identifying the important ones is extraordinarily difficult. Katie Pollard’s team is developing AI models that predict which DNA changes are most likely to alter cell function. She collaborates with the labs of Seth Shipman, who can make thousands of DNA edits simultaneously, and Vijay Ramani, whose technologies measure the effects of the edits. Together, they are creating an iterative system designed to eventually help scientists interpret a person’s genome and identify opportunities for prevention or treatment.

Christina Theodoris develops foundation models that learn how genes work together across millions of individual cells. Her first model, Geneformer, can predict how changing the activity of a gene will affect a cell and has already revealed potential drug targets for heart disease. Her lab is also developing models such as MaxToki to study how cells change throughout aging. Rather than testing an astronomical number of gene combinations in the laboratory, these models allow scientists to perform experiments virtually, prioritize the most promising ideas, and uncover biological relationships they may not otherwise have considered.

Steve Finkbeiner’s lab has invented a thinking microscope that combines robotics with reinforcement learning. Given a goal—such as making a diseased neuron behave more like a healthy one—the microscope can design experiments, manipulate individual cells, observe what happens, learn from the results, and choose what experiment to perform next. Thousands of experiments can happen simultaneously. The team aims to use this autonomous system to better understand the cause of neurodegenerative diseases like Alzheimer’s disease, Parkinson’s disease, and ALS.

Many families experience Alzheimer’s disease across generations without knowing what genetic changes are driving their risk. Ryan Corces and his team are using AI to search across millions of possible genetic changes and narrow them to a manageable set that scientists can test experimentally. The goal is to uncover previously unknown causes of Alzheimer’s and other neurodegenerative diseases, help identify new therapeutic targets, and eventually provide families with clearer information about the biology underlying their disease.

Yadong Huang’s lab is adapting Geneformer to build AI models specifically focused on Alzheimer’s disease. By training models on large collections of human and mouse single-cell data, the team aims to identify the relatively small number of gene changes that actually drive disease progression. Huang’s group is also developing what he calls “virtual pharmacology:” an AI model trained on experimental data showing how thousands of compounds affect human brain cells. Their goal is to identify existing drugs that could reverse disease-related changes and accelerate the path toward promising treatments for Alzheimer’s.

The earliest effects of Alzheimer’s can begin long before obvious memory problems appear. Jorge Palop’s team used a deep-learning tool to analyze videos of mice and identify subtle patterns of spontaneous behavior that conventional tests could miss. The model detected increasingly disorganized behavior as disease progressed, providing researchers with a more sensitive way to measure early brain dysfunction. Similar approaches could eventually help researchers analyze behavior in people using ordinary video, potentially offering new ways to detect neurological disease earlier and measure whether treatments are working.

Alex Marson’s lab uses CRISPR to make thousands of different genetic changes in immune cells and test whether those changes make the cells better at fighting cancer. In collaboration with Barbara Engelhardt, the team is developing computer-vision tools that extract far more information from videos of these experiments. Rather than simply counting how many cancer cells remain, AI can track individual immune cells and characterize behaviors such as how they move, interact, and kill. These insights could help scientists engineer safer, more effective cancer immunotherapies, particularly for tumors that do not respond to existing treatments.

Wearable devices can continuously measure heart rate, temperature, sleep, activity, and other signals, creating an unprecedented opportunity to study how the body changes throughout the menstrual cycle. But these data are noisy, incomplete, and difficult to interpret. Barbara Engelhardt’s lab develops AI and statistical methods specifically designed to make sense of these complex measurements. By combining wearable data with information about hormones, symptoms, reproductive conditions, and other factors, her team aims to uncover patterns that have been missed by traditional studies and ultimately give people more personalized information about their health.

As antibiotics increasingly fail to treat bacterial infections, phages (viruses that naturally kill bacteria) provide an exciting alternative. Through Gladstone’s Center for PhAIge Therapy, Seth Shipman, Katie Pollard, Melanie Ott, and Sukrit Silas are combining high-throughput experiments with AI to learn the rules governing interactions between phage and bacteria. Their goal is to predict which phage will kill a patient’s particular infection—and eventually design effective phages when the right one does not already exist—making phage therapy faster, more reliable, and more widely available.

Nadia Roan and Melanie Ott are helping lead an international effort to create a comprehensive atlas of how HIV affects the immune system. The researchers will use AI to integrate complex, single-cell datasets and compare cells from people with HIV with existing data from people never exposed to the virus. The resulting open-access atlas is intended to reveal how HIV reshapes immunity and characterize cells where the virus hides. These insights could guide better therapies, address long-term health complications, and advance efforts toward an HIV cure.

Modern experiments can profile thousands or millions of individual cells, but different studies may label and categorize those cells in different ways, making comparisons difficult. Katie Pollard’s team developed CellWalker2, a computational tool that recognizes relationships between cell types and connects data across experiments, tissues, and even species. By understanding that some cell types are closely related while others are fundamentally different, the tool can organize previously disconnected datasets and reveal biological relationships that researchers might otherwise miss—helping scientists extract more knowledge from the enormous volume of single-cell data now being generated.

Scientists long thought DNA wrapped around structures called nucleosomes was essentially inaccessible until it became unwrapped. Vijay Ramani and his collaborators used an AI-powered computational method to analyze detailed DNA sequencing data and discovered a much more dynamic picture. The model detected subtle structural differences showing that DNA can remain partially accessible even while wrapped around nucleosomes. This finding reveals a previously unrecognized layer of gene regulation and could help explain how small changes in DNA organization contribute to cancer, aging, and other complex diseases.

Andrew Yang’s lab uses AI to understand the blood-brain barrier—and turn it from an obstacle into a therapeutic opportunity. By combining experimental data with AI models that predict protein structures and interactions, his team is identifying new routes for delivering medicines into the brain. Yang also collaborates with Christina Theodoris to apply AI to molecular data spanning the human lifespan, pinpointing potential targets that could slow or reverse aspects of brain aging and ultimately help prevent age-related neurological disease.