Can we predict T cell specificity with digital biology and machine learning?

Abstract

Recent advances in machine learning and experimental biology have offered breakthrough solutions to problems such as protein structure prediction that were long thought to be intractable. However, despite the pivotal role of the T cell receptor (TCR) in orchestrating cellular immunity in health and disease, computational reconstruction of a reliable map from a TCR to its cognate antigens remains a holy grail of systems immunology. Current data sets are limited to a negligible fraction of the universe of possible TCR–ligand pairs, and performance of state-of-the-art predictive models wanes when applied beyond these known binders. In this Perspective article, we make the case for renewed and coordinated interdisciplinary effort to tackle the problem of predicting TCR–antigen specificity. We set out the general requirements of predictive models of antigen binding, highlight critical challenges and discuss how recent advances in digital biology such as single-cell technology and machine learning may provide possible solutions. Finally, we describe how predicting TCR specificity might contribute to our understanding of the broader puzzle of antigen immunogenicity.

Antigen‐specific and cross‐reactive T cells in protection and disease

“…Computational prediction of TCR specificity can have immense utility in furthering our understanding of systems immunology and would lead to breakthroughs in translational immunotherapies. Recent reviews thoroughly highlight the various in silico modeling approaches aimed at predicting peptide–MHC binding, cross‐reactivity, immunogenicity, and TCR–pMHC interaction 221–226 . Despite recent advances in both high‐throughput TCR–antigen discovery and machine learning approaches, there exist significant challenges that need to be addressed.…”

The pitfalls of negative data bias for the T-cell epitope specificity challenge

“…Using high-throughput sequencing technologies, it is possible to map the sequences of the TCRs within a biological sample, for example from blood of an individual with a specific disorder. However, the number of possible TCR sequences is incredibly large, with a conservative estimate in the range of 10 15 unique sequences 6 . Consequently, the epitope targets of the vast majority of TCRs are unknown.…”

“…Therefore, zero-shot TCR-epitope annotation-i.e. predicting TCR-epitope binding for novel, unseen epitopes-is currently seen as the 'holy grail' of immunology 6 . This requires machine learning methods to actually learn the underlying recognition code of the TCRs, which has turned out to be a substantially harder problem.…”

Reply to: The pitfalls of negative data bias for the T-cell epitope specificity challenge

“…efforts to raise this point for T-cell epitope specificity modeling, which is known clearly by the community that different negative data sampling strategy will influence the prediction results 2,3 . Therefore, proper negative data sampling strategy should be carefully selected, and this is exactly what PanPep has noticed, emphasized and performed 4 .…”

“…This is also well known for other communities besides TCR-peptide recognition, such as for protein-ligand binding prediction 6 etc. However, in the TCR-peptide recognition community, it is well known that the cross-reactivity of TCRs is a significant challenge in the TCR-peptide recognition problem and it can pose difficulties for many models in the pre-processing stage 2 . Experimental methods can have a high false-negative rate 1 , resulting in many potential cross-reactivity in existing known binding TCRs, making PU Learning more challenging.…”