Quantitative approaches for decoding the specificity of the human T cell repertoire

Abstract

T cell receptor (TCR)-peptide-major histocompatibility complex (pMHC) interactions play a vital role in initiating immune responses against pathogens, and the specificity of TCRpMHC interactions is crucial for developing optimized therapeutic strategies. The advent of high-throughput immunological and structural evaluation of TCR and pMHC has provided an abundance of data for computational approaches that aim to predict favorable TCR-pMHC interactions. Current models are constructed using information on protein sequence, structures, or a combination of both, and utilize a variety of statistical learning-based approaches for identifying the rules governing specificity. This review examines the current theoretical, computational, and deep learning approaches for identifying TCR-pMHC recognition pairs, placing emphasis on each method's mathematical approach, predictive performance, and limitations.

Keywords: TCR; binding prediction; deep learning; machine learning; pMHC; protein-protein interaction.

Figure 1

Modeling approach to TCR-pMHC prediction based on input data. (A) Models trained purely on TCR and peptide sequence input data feature multiple sequence alignment (MSA) on input data matrices to identify patterns, followed by identification of potential interaction pairs using various algorithms techniques. (B) Models trained on input structural data models commonly aim to identify the TCR-pMHC binding interface, along with associated information on binding affinity. Predictions are often made by determining similarity in secondary structure in the interfacial region of the binding interface, from which physical parameters like binding affinity and kinetic data (K_D , K_off ) can be estimated. (C) A third ‘hybrid’ category of inferential model synergistically combines sequence and structural data in the training step.

Figure 2

List of commonly used inference-based models of TCR-pMHC specificity partitioned by approach: sequence-based, structure-based, and hybrid.

Figure 3

A schematic illustration of various deep learning architectures employed for TCR-pMHC interaction prediction: (A) 2D CNN-based prediction of TCR-pMHC interactions: The pairwise features of protein sequences are encapsulated in a 2D matrix representation, which serves as input for the 2D CNN. The CNN systematically samples the entire protein pairwise feature space, processing the data to facilitate the learning of TCR-pMHC interactions, (B) RNNs utilize auto-regressive learning to generate sequences, which can be applied in the context of TCR-pMHC interaction prediction, (C) In the GAN framework, a mapping from a prior distribution to the design space can be obtained through adversarial training, enabling the generation of novel TCR-pMHC interaction predictions, (D) VAEs can be jointly trained on protein sequences and their properties to construct a latent space that correlates with the properties of interest, for example, the TCR binding capacity of unevaluated target peptides.