T cell breakthrough a step towards personalised medicine to prevent disease

In a major breakthrough, scientists from Monash University have unlocked one of immunity medicine’s biggest problems – how to accurately predict which T cells and their receptors work, at a molecular level, to prevent disease.

The team from Monash’s Biomedicine Discovery Institute (BDI) – with Shanghai Jiao Tong University and Yale University – used a custom artificial intelligence (AI) modelling in their collaboration, which furthers the quest towards personalised cellular and genetic medicine.

A T cell receptor, one of the important pieces in this puzzle, is known to the scientists as a TCR. It’s a web of proteins that get the cells active when an antigen (such as a bacteria or virus, or any “toxin”) is detected, triggering an immune response.

The problem until now has been the huge range of them, and also the similarly huge range of the epitopes, inside an antigen, which the receptors bind to.

The research is now published in Nature Machine Intelligence. Lead senior author Professor Jiangning Song, who heads the AI-driven Bioinformatics and Biomedicine Laboratory at BDI, explains that understanding how T cells identify antigens is crucial in the development of better vaccines, immunotherapies, and treatments for cancers, autoimmune diseases and infectious diseases.

Professor Song’s work straddles biomedicine and computing, as a member of the Monash Data Futures Institute, the Australian Research Council’s Centre of Excellence in Advanced Molecular Imaging, and Monash’s Alliance for Digital Health.

The impact of EPACT

The team’s paper, in its introduction, says despite “extensive research efforts”, predicting the TCR and antigen binding pairs was a significant challenge, but the AI deep-learning tool called EPACT, looking at the epitopes in the antigens and the specific CD8+ T cells, was successful.

CD8+ cells work in immune defence against viruses and bacteria, and in tumour surveillance, killing infected cells.

The researchers applied the AI method to COVID SARS-CoV-2, the relevant T cells with the predicted binding strength aligning well with immune responses after vaccination. The T cell receptors activate the cells via specific antigens shown by MHC class 1 molecules.

“A major hallmark of this specific program is that we needed to deal with an avalanche of high-dimensional multimodality biomedical datasets,” says Professor Song.

“The datasets come from sequencing data, single-cell transcriptomic data, 3D structure data, interaction data. In the end, we needed to think how we can better develop an approach that can enable us to achieve much better understanding across this type of data. Can we really use that on undiscovered knowledge, and turn it into something that can become translational, for example.”

A dedicated AI model

One of the issues in this type of science is what professor Song calls the “widening gap” between widely-available cell sequencing data and increasingly expensive 3D cell structure data to load into the emerging AI tools.

“Experimental methods that map TCR sequence data to antigen specificity face several issues, including high cost, technical complexity, and limited coverage. Existing computational methods have difficulties in predicting binding TCRs for unseen epitopes due to the lack of high-quality data.

“This is why we collaborated to develop a more dedicated, specialised AI model, published in this paper,” he says. “This is just the first step is to develop a better AI approach that will allow researchers to design a safer, more efficacious, more effective T cell receptor that can achieve individualised immunotherapy, for example, from predicative AI to generative AI. That's the long-term goal.

Read more: Breakthrough heralds immune-personalised medicine for lupus

One of the main findings in the study was to do with T cell residue interactions with epitopes in the antigen – “a critical development”, he says.

“A lot of methods cannot achieve residue-level interaction prediction, pinpointing and identifying critical residue that’s responsible for not just the binding affinity, but also for the cross-reactivity. Our model can be further leveraged to guide the epitope-like vaccine in the future.”

Revealing ‘multi-scale’ interactions

The primary finding, he says, from a long and complex molecular investigation such as this, is that an AI model now exists that can show “multi-scale” interactions between the T cells, molecules in cells, peptides in epitopes.

“It can also identify antigen-specific T cell clusters, corroborate SARS-CoV-2 spike-specific T cell response upon vaccination.”

Yumeng Zhang, the first author of this study and a PhD student in Professor Song's BDI lab, implemented the AI model. “Our model can also be extendable to identify immunogenic neoantigens and design individualised biomarkers”.

Professor Song says the AI model developed by the research team is a “powerful computational tool with great potential of being applied to accelerate the assessment of TCR-based immunotherapies and vaccines in clinical studies”.