epiTCR: a highly sensitive predictor for TCR–peptide binding

Abstract

Motivation Predicting the binding between T-cell receptor (TCR) and peptide presented by HLA molecule is a highly challenging task and a key bottleneck in the development of immunotherapy. Existing prediction tools, despite exhibiting good performance on the datasets they were built with, suffer from low true positive rates when used to predict epitopes capable of eliciting T-cell responses in patients. Therefore, an improved tool for TCR-peptide prediction built upon a large dataset combining existing publicly available data is still needed. Results We collected data from five public databases (IEDB, TBAdb, VDJdb, McPAS-TCR, and 10X) to form a dataset of > 3 million TCR-peptide pairs, 3.27% of which were binding interactions. We proposed epiTCR, a Random Forest-based method dedicated to predicting the TCR-peptide interactions. epiTCR used simple input of TCR CDR3β sequences and antigen sequences, which are encoded by flattened BLOSUM62. epiTCR performed with AUC (0.98) and higher sensitivity (0.94) than other existing tools (NetTCR, Imrex, ATM-TCR, and pMTnet), while maintaining comparable prediction specificity (0.9). We identified seven epitopes that contributed to 98.67% of false positives predicted by epiTCR and exerted similar effects on other tools. We also demonstrated a considerable influence of peptide sequences on prediction, highlighting the need for more diverse peptides in a more balanced dataset. In conclusion, epiTCR is among the most well-performing tools thanks to the use of combined data from public sources and its use will contribute to the quest in identifying neoantigens for precision cancer immunotherapy. Availability epiTCR is available on GitHub (https://github.com/ddiem-ri-4D/epiTCR). Supplementary information Supplementary data are available at Bioinformatics online.

TCR-H: Machine Learning Prediction of T-cell Receptor Epitope Binding on Unseen Datasets

“…For this, the AUC of the ROC of TCR-H model is 0.91 (Table 2) which is comparable to existing methods. In particular, it is equal to the performance of an SVM model built using sequences encoded using BLOSUM62 for which the resulting matrices were transformed into feature vectors through flattening and concatenation (Pham et al, 2023).…”

“…With the increasing availability of TCR and epitope sequencing data from high-throughput sequencing techniques in publicly available data resources (Jokinen et al, 2021), AI and ML approaches have been able to be used to predict TCR CDR3β binding to epitopes presented by MHC class 1 (MHC-I) (Hudson et al, 2023). These tools apply methods ranging from relatively simple ML models such as Random Forest and clustering (Chronister et al, 2021; Dash et al, 2017; Jokinen et al, 2021; Pham et al, 2023) to various forms of deep learning-based AI techniques, including convolutional and recurrent neural networks (Bravi et al, 2023; Cai et al, 2022; Darmawan et al, 2023; Gao, Gao, Li, et al, 2023; Jiang et al, 2023; Jokinen et al, 2023; Myronov et al, 2023; D. Wang et al, 2023; Z.…”

“…The choice of supervised ML over deep learning was motivated in part by the ability of the model to be based on explicit features posited to be important for binding, which in turn may aid future model interpretation, explainability and rationality. Furthermore, it has been shown that the use of features that lend themselves to ML models leads to better performance as well (Pham et al, 2023). Most ML models have been sequence based, encoding sequences using either the BLOSUM substitution matrix (Cai et al, 2022; Darmawan et al, 2023; Gao, Gao, Li, et al, 2023; Pham et al, 2023) and/or Atchley factors(Atchley et al, 2005).…”

xTrimoPGLM: Unified 100B-Scale Pre-trained Transformer for Deciphering the Language of Protein

“…The classification task is to predict whether a given paired TCR sequence and peptide can bind or not. The evaluated data is major from VDJdb, processed and curated from Pham et al (87).…”

TSpred: a robust prediction framework for TCR-epitope interactions based on an ensemble deep learning approach using paired chain TCR sequence data

“…Deep learning-based methods use a wide variety of architectures, such as the convolutional neural networks (CNNs) (9)(10)(11)(12), long short-term memory (LSTM) (13), autoencoder (13), and attention mechanism (6,14,15). Some tools, such as TITAN (6), ImRex (12), pMTnet (16), epiTCR (17), and TEINet (18), consider only the information of the TCR beta chain. Other methods, such as ERGO (13), MixTCRpred (15), NetTCR-2.0 (11), NetTCR-2.1 (10), and NetTCR-2.2 (9), take into consideration the paired alpha and beta chain information.…”