Evolutionary Transfer Learning to Chart the Pathophysiology of Human Developmental Disorders
A single fertilized cell becomes an entire animal: heart, brain, bones, and all, in just weeks (mouse) or months (human). This everyday miracle of development is one of the most remarkable processes in nature, yet remains poorly understood. We don’t have a complete “movie” of how cells multiply, change identity, and organize themselves into tissues and organs. When something goes wrong along the way, the result can be a lifelong developmental disorder.
Every year, millions of children are born with developmental disorders affecting the heart, brain, skeleton, and other organs. For many of these conditions, scientists know which gene is broken, but not how it causes disease. The link between genes and symptoms is rarely one-to-one: a single gene can affect completely different parts of the body, while different genes can produce strikingly similar symptoms. These patterns suggest that disease genes are connected through shared developmental pathways. But because they have been studied gene-by-gene, the connections remain hidden.
We cannot study human development directly, since early embryos and fetuses are largely inaccessible. But animals like mice, zebrafish, and monkeys develop in ways that closely mirror our own, and scientists have already mapped their development cell by cell, in extraordinary detail. Animals share deep similarities with humans: in DNA sequences, in the molecular switches that turn genes on and off, and in how cells become specialized tissues. Just as ChatGPT learns patterns from billions of sentences, I will develop AI to learn patterns from millions of cells across many species, and then predict what happens to a human cell at any point in development, even though we cannot directly observe it. The result will be a first-of-its-kind virtual movie of human development, from conception onward.
With this movie, we can place disease genes side by side and ask which ones disrupt the same cells, stages, or pathways. Instead of solving each disorder from scratch, we can use what is known about well-studied genes to illuminate poorly understood ones. The payoff for children is concrete: earlier diagnoses, better predictions of complications, and treatments that could help many disorders at once.