Cold Spring Harbor Laboratory Study Applies AI Concepts to T Cell Training

Cold Spring Harbor Laboratory announced in a press release a study that applies machine learning concepts to how T cells are trained in the thymus. Researchers used single cell sequencing and AI simulations to estimate that, during thymus training, each T cell interacts with about 240 cells that present antigens out of a random sample of 2,000.

The study focuses on negative selection, the process that removes T cells that bind to the body's own protein fragments. The researchers found that T cells can still recognize other self peptides through generalization, a concept also used in machine learning.

The team reported two conditions that support this process. The abundance of self peptides in the thymus closely matches their abundance in tissues across the body, and T cell receptors can recognize several similar peptides. The study found that this process can correctly delete 90% of self reactive T cells even when each one encounters only 10% of the body's self peptides.

The researchers also tested whether failures in this process could relate to autoimmunity. Their AI model reproduced features of autoimmune polyendocrine syndrome type 1, a rare autoimmune disease.