NetTCR: sequence-based prediction of TCR binding to peptide-MHC complexes using convolutional neural networks

Abstract

Predicting epitopes recognized by cytotoxic T cells has been a long standing challenge within the field of immuno-and bioinformatics. While reliable predictions of peptide binding are available for most Major Histocompatibility Complex class I (MHCI) alleles, prediction models of T cell receptor (TCR) interactions with MHC class I-peptide complexes remain poor due to the limited amount of available training data. Recent next generation sequencing projects have however generated a considerable amount of data relating TCR sequences with their cognate HLA-peptide complex target. Here, we utilize such data to train a sequence-based predictor of the interaction between TCRs and peptides presented by the most common human MHCI allele, HLA-A*02:01. Our model is based on convolutional neural networks, which are especially designed to meet the challenges posed by the large length variations of TCRs. We show that such a sequence-based model allows for the identification of TCRs binding a given cognate peptide-MHC target out of a large pool of non-binding TCRs.

Current challenges for epitope-agnostic TCR interaction prediction and a new perspective derived from image classification

“…However, recent studies have shown that while generic TCR-epitope prediction models can achieve a reasonable performance on interactions that involve epitopes already encountered training, they cannot yet reliable extrapolate to novel epitopes (i.e. epitope-agnostic models) (15,20). This is likely caused by the large heterogeneity in potential TCR and epitope sequences, and the limited amount of readily available training data that covers this sequence space.…”

Understanding Adaptive Immune System as Reinforcement Learning

“…We can also introduce pretraining to mimic and analyze the thymic selection [56]. Furthermore, if comprehensive quantitative data on the interactions between antigens and Th clones are obtained by future measurement technologies [57][58][59][60], we can include that information on the weights of the network.…”

Prediction of specific TCR-peptide binding from large dictionaries of TCR-peptide pairs

“…ERGO is based on LSTM networks to encode sequential data. Previous models by Jurtz et al (18) used convolutional neural networks (CNN) for the similar task. While CNN are good at extracting position-invariant features, RNN (in particular LSTM) can catch a global representation of a sequence, in various NLP tasks (36).…”

“…Finally, we compared ERGO's performance in a mixed setting of TCRs of unknownspecificities recently proposed by Jurtz et al (18), who used a convolutional neural network (CNN) based model, NetTCR, for predicting binding probability of TCR-HLA-A*02:01 restricted peptide pairs. Jurtz et al experimented on two datasets, one downloaded from IEDB was used to train the model and an additional dataset, generated using the MIRA assay provided by Klinger et al (30), was used for evaluating the model, by testing the model performance on shared IEDB and MIRA peptides ( Table 2 and Table 3) Table 2).…”

“…Important steps have been made in this direction by Glanville et al (4) and Dash et al (16), who detected the clear signature of short amino acid motifs in the CDR3 region of TCRβ and TCRα in response to specific peptides presented by specific MHC molecules. This work was then extended by two more recent efforts that combined such motifs with machine learning approaches (17,18) to predict peptide specific TCRs vs. naïve TCRs, using either Convolutional Neural Networks (18), or Gaussian Processes (17). These methods significantly outperform random classification in practically all tested datasets.…”