The Human Cell Is Wildly Complex. Can AI Decode It? | Silvana Konermann | TED

Silvana Konermann, driven by a childhood fascination with biology and a personal sense of purpose, is tackling complex diseases like Alzheimer’s by combining CRISPR, single-cell sequencing, and AI to build a “universal virtual cell.” Because Alzheimer’s arises from unique combinations of genetic and environmental factors in each patient, the key is finding shared targets that can convert diseased cells back to healthy states. Her lab is generating massive datasets—over a billion perturbational experiments—to teach AI the “language of the cell” through RNA expression, enabling predictions of which gene or chemical interventions will work. A first state-of-the-art model is already released, though it still needs far greater accuracy to be truly useful; safety is built in by restricting it to human cells. With a team of over 300 at Arc, she envisions AI and biology united under one roof, predicting that within four to five years such models will transform medicine for currently untreatable diseases.

• ▶ 0:14 Silvana Konermann grew up in a small Swiss town with non-scientist parents, but her early fascination with nature and biology drove her to push her way into a lab as a teenager.
• ▶ 0:40 Her first science project won national and European Union competitions, giving her the confidence to keep pursuing science — though it also created a lasting sense of responsibility to do something meaningful.
• ▶ 1:39 Since her undergraduate years, she has focused on Alzheimer's disease, struck by the fact that scientists knew the late-stage brain changes but had no idea how the disease begins or how to treat it.
• ▶ 2:19 Alzheimer's is a "complex disease": not just complicated, but driven by many different risk factors, with essentially every patient having a unique combination — unlike a single-cause infection.
• ▶ 3:04 Because these diseases combine genetic changes and environmental factors, the central challenge is finding a shared target among diverse patients that could be used to treat and fix the disease.
• ▶ 3:54 Recent convergence of single-cell sequencing, CRISPR gene editing, and AI makes it possible to tackle complex diseases.
• ▶ 5:19 Treating RNA as the “language of the cell” is key: AI can learn from dynamic RNA expression just as it learned human language.
• ▶ 6:56 Building a useful model requires generating massive new biological datasets—especially precise single-cell measurements that capture how cells respond to perturbation.

• ▶ 7:57 Konermann explains that experiments combine CRISPR-based targeted genome changes (switching genes off/on one cell at a time) with single-cell RNA sequencing to measure the effect of each perturbation.

• ▶ 8:28 The project targets at least one billion physical, biological experiments over four years, with about 60 million already completed; at ▶ 9:00 she notes barcoding and pooling make this scale feasible.

• ▶ 10:18 Using the model, they can ask what genetic or chemical intervention would convert diseased cells (e.g., Alzheimer's microglia) back to healthy ones—addressing the guess-and-check bottleneck described at ▶ 11:03.

• ▶ 11:28 The universal virtual cell must generalize to new cell types or disease states it has never seen in training data — a key challenge driving experiment design.
• ▶ 12:08 A first version of the model is already released and is state-of-the-art, but still far from the accuracy needed to be truly useful; a "state designer" interface lets users specify a target cell state and get likely modifications.
• ▶ 13:47 Safety is addressed by restricting the model to human cells only, making it difficult to abuse, while the same tool could provide a rapid-response defense by revealing how a virus targets genes in specific cells.
• ▶ 15:16 Arc has grown to over 300 people in just one year, despite launching only four years ago.
• ▶ 15:20 A core vision is uniting AI and biology under one roof, with a focus on applying machine learning to biology.
• ▶ 16:23 Konermann predicts a totally different way of doing biology, with models accurate enough to transform medicine for diseases like Alzheimer's within four to five years.

• ▶ 0:04 Origins and the Alzheimer's Challenge (0:04 - 3:27) - Covers Konermann's early inspiration, her prodigy path, and the difficulty of tackling complex diseases like Alzheimer's.
• ▶ 3:27 A New Approach: AI, RNA, and Data (3:27 - 7:52) - Explains the three converging technologies, treating RNA as a language, and the need for massive biological data.
• ▶ 7:52 Building and Scaling the Experiment Engine (7:52 - 11:28) - Describes CRISPR perturbations, single-cell measurement, barcoding, millions of experiments, and disease-state prediction.
• ▶ 11:28 The Universal Virtual Cell and Safety (11:28 - 14:56) - Introduces the virtual cell model, its public release, safety concerns, and potential defense against viruses.
• ▶ 14:56 Team and Vision for Medicine (14:56 - 17:04) - Showcases the Arc team's growth and the mission to transform medicine for Alzheimer's and other diseases.