Perturb-seq

A cell uses thousands of genes. A gene is a piece of DNA containing the blueprint for one protein, and genes switch each other on and off. Perturb-seq is a way of finding out which gene does what: researchers switch off a gene and look at what then happens to the rest.

How does it work?

You can compare it to a large factory in which you shut down one machine at a time. Then you walk through all the departments and see what changes. That is how you learn which machine is needed for what.

The name says what happens. Perturb means to disturb: in a cell, a gene is switched off, using a technique from the CRISPR family. Seq stands for sequencing, reading out the genetic activity of that cell. For each cell, researchers therefore measure which other genes start to work harder or less hard because of the disturbance.

What is special is the scale. The same measurement is made in large numbers of cells at once, each with a different gene switched off. The result is a map of cause and effect: which gene controls which other gene. Such a map yields an enormous amount of data, and working out what all those changes mean takes a great deal of time.

Why it matters

Anyone who knows which genes work together has a better understanding of how diseases arise and where research into medicines can begin. Perturb-seq provides the raw map for that.

An experiment on mouse brains showed how AI can help to read such a map. The company PerturbAI switched off 1,947 disease-related genes in the brains of mice. OpenAI then put almost 2,000 AI agents to work at the same time on that large dataset of mouse brains, an atlas of 7.7 million brain cells. Each agent took one gene. In 32 minutes they came up, together, with explanations and new research ideas, work that would take people months (link). Those ideas are hypotheses: they now have to be tested in the lab.