Machine rapidly sorts cells after learning their shape

A new technique can quickly sift through thousands of cells, isolating only those of interest¹.

Autism scientists could use the approach to separate different kinds of brain cells — microglia from neurons, or one type of neuron from another, for example.

Methods for rapidly sorting large numbers of cells require researchers to first label only the cells of interest with fluorescent molecules. They can then sort cells based on the presence or absence of fluorescence.

The new technique separates cells based on their size and shape, circumventing the need to tag cells ahead of time.

The researchers first stain all the cells in a mixture with a fluorescent molecule. They then train a custom machine-learning algorithm to recognize the light-wave pattern emitted by the target cell type.

A detector in a modified cell-sorting device captures the light patterns and transmits them to the algorithm. When the program detects the cell type of interest, it signals to another component of the device, which drops the cell into a separate container.

The researchers trained their algorithm to detect certain cancer cells and other blood cells. The method then sorted these cancer cells from the blood cells at a rate of 3,000 cells per second. The team also used the technique to sort pancreatic cancer cells from breast cancer cells. They published the results in June in Science.

The researchers are working on ways to generate a cell-specific signature without using fluorescent stains. In the meantime, they have founded a company called ThinkCyte to make the cell-sorting device commercially available.