New AI method could aid in development of T cell-based cancer therapies

A team at the University of Texas (UT) Southwestern, US, have developed an artificial intelligence (AI) technique that can identify which cell surface peptides produced by neoantigens are recognised by the immune system. The method, known as the pMTnet technique, could lead to new ways to predict cancer prognosis and potential responsiveness to immunotherapies which may help develop treatments such as cancer vaccines and T-cell based therapies.

“For the immune system, the presence of neoantigens is one of the biggest differences between normal and tumour cells,” explained Tianshi Lu, first co-author. “If we can figure out which neoantigens stimulate an immune response, then we may be able to use this knowledge in a variety of different ways to fight cancer.”

The team trained a deep learning-based algorithm that they named pMTnet using data from known binding or non-binding combinations of three different components: neoantigens; major histocompatibility complexes (MHCs) that present neoantigens on cancer cell surfaces; and the T cell receptors (TCRs) responsible for recognising the neoantigen-MHC complexes. They then tested the algorithm against a dataset developed from 30 different studies that had experimentally identified binding or non-binding neoantigen T-cell receptor pairs. The experiment showed that the new algorithms had a high level of accuracy.

The researchers used this new tool to gather insights on neoantigens catalogued in The Cancer Genome Atlas, a public database that holds information from more than 11,000 primary tumours. pMTnet showed that neoantigens generally trigger a stronger immune response compared with tumour-associated antigens. It also predicted which patients had better responses to immune checkpoint blockade therapies and had better overall survival rates. 

“As an immunologist, the most significant hurdle currently facing immunotherapy is the ability to determine which antigens are recognised by which T cells in order to leverage these pairings for therapeutic purposes,” commented corresponding author Alexandre Reuben. “pMTnet outperforms its current alternatives and brings us significantly closer to this objective.”

The study was published in Nature Machine Intelligence.