NL
DE
An amazing variety of T cells protect us from harmful pathogens: our immune system can make more different T cells than there are grains of sand on the earth or stars in the universe. Modern technologies allow us to sequence T cell receptor repertoires, generating rich and important data that is very difficult to analyse.
In this research, Johannes Textor, Franka Buytenhuijs (RU) and Judith Mandl (McGill) asked a simple but fundamental question: can we predict how strongly a T cell will react to its antigen just by looking at the sequence of its receptor?
Together with team members from the groups of Jürgen Westermann (University of Lübeck), they answered this question by collecting millions of T-cell sequences and training a machine learning algorithm to distinguish strongly- from weakly-reacting T cells. Using this algorithm, they found that using hydrophobic amino acids and certain gene fragments makes some T cells bind more strongly.
This work is an example of how researchers can use machine learning to extract fundamental insights from large and complex datasets. A particularly important step was performed by Heather Melichar's group (McGill), who tested predictions of the machine learning algorithm in genetically engineered mice expressing sequences not used during training.
The team hopes that the results of this research will help to make T-cell-based vaccines and other immunotherapies more effective.
The results have been published in Cell Systems on 20 December and are featured on the journal's cover.
Photo
Top left to right; Johannes Textor (Radboudumc), Franka Buytenhuijs (RU), Dakota Rogers (McGill)
Bottom left to right; Heather Melichar (McGill), Jürgen Westermann (Uni Lübeck), Judith Mandl (McGill)
