Motif in T cells for the Prediction of INTeractions

Project Information

MT-PoINT

Grant agreement ID: 101027973

DOI

10.3030/101027973

Project closed

EC signature date 7 April 2021

Start date 1 July 2021

End date 30 June 2023

Funded under

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Total cost

€ 191 149,44

EU contribution

€ 191 149,44

Coordinated by

UNIVERSITE DE LAUSANNE
Switzerland

Periodic Reporting for period 1 - MT-PoINT (Motif in T cells for the Prediction of INTeractions)

Reporting period: 2021-07-01 to 2023-06-30

Summary of the context and overall objectives of the project

T cells, an essential part of our cellular immune response, have a critical responsibility in defending our bodies against harmful invaders such as viruses, bacteria, and even cancer cells. By continuously patrolling our body, they can recognize molecular fragments presented by infected or cancerous cells (the so-called epitopes) and initiate a response to eliminate them. In the field of cancer immunotherapy, new innovative treatments involve engineering or stimulating specific T-cells in the laboratory and then injecting them into cancer patients in order to target and destroy tumoral cells. These groundbreaking therapies have revolutionized cancer treatments, offering long-term advantages to a considerable portion of patients, including those with advanced-stage cancers. Despite considerable research, our understanding of the exact mechanisms by which T cells identify and eradicate cancerous or infected cells remains incomplete, hindering the developments of treatments that utilize or modify T cells to selectively target particular cells.
The ultimate goal of the MT-PoINT project has been to develop computational models that can predict if a given T cell would recognize epitopes presented by cancerous or infected cells, and eventually eliminate them. Recent years have witnessed the emergence of a variety of immune assays that allow to isolate and sequence T cells paired with the cognate epitopes. Combining publicly available and in-house generated data with innovative algorithmic developments, we developed TCRpred: a deep-learning software that takes the T cell and the epitope sequence as inputs and predicts the likelihood of their interaction. Through a comprehensive analysis, we evaluated the precision and limitations of our tool, and used it to gain deeper insights into T cell-epitope recognition mechanism. TCRpred can serve as a valuable tool for scientists, enabling them to gain deeper insights into T cell-mediated immunity and, in turn, greatly expedite advancements in cancer immunotherapy clinical research and development.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

The research work was conducted via 2 work packages (WPs). WP1 consisted of gathering and curating an extensive dataset of T cell-epitope sequence data and developing a computational predictor. This effort resulted in the acquisition of approximately 17,000 T cell sequences targeting numerous viral and cancer epitopes. We then utilized this extensive dataset to develop T-epitope interaction predictors. Our exploration encompassed diverse approaches, ranging from distance-based classifiers to more advanced machine learning or deep learning models. Ultimately, we developed TCRpred, a sequence-based deep-learning model, which can accurately predict T cell - epitope interactions with higher accuracies with respect to existing tools. Furthermore, we established TCRpred as a valuable quality control tool for TCR-sequencing datasets of T cells isolated using DNA-barcoded pMHC multimers, a technology expected to provide a large amount of T cell -epitope data in the near future. Leveraging TCRpred, we also gained improved insights into the specificity of dual α T cells, a sub-population comprising up to 10% of the T cell immune system, which has been observed for over three decades, but still remains poorly understood. Finally, we employed TCRpred to analyze T cell responses in COVID-19 patients. Our predictions revealed an enrichment of T cells specific to an immunodominant SARS-CoV-2 epitope in COVID-19-positive patients.
In order to pave to new computationally-guided treatments for cancer, it was crucial to access the TCRpred predictive performance for predicting T cell - cancer epitope interactions and eventually validation in a clinically relevant setting (WP2). We collaborated with other scientists at UNIL to generate additional data for epitopes that hold significant relevance in cancer immunotherapy. Through a collaboration with Dott. Dunn, we employed a phage display experiment to generate over 70,000 distinct T cells specific to the cancer epitope called NY-ESO-1. Furthermore, in collaboration with Dr. Arber, we sampled T cells targeting NY-ESO-1 directly from patients. While investigation is still ongoing, the data collected through collaborations with Dr. Dunn and Dr. Arber allowed us to extensively test the accuracy of TCRpred to predict the T cells targeting the cancer epitope NY-ESO-1, with the future aim of prioritizing T cells candidates for cancer immunotherapy.

The main manuscript which describes TCRpred predictive performance and applications is currently under review. The TCRpred tools and applications were presented in 6 different conferences, 2 poster sessions. Furthermore, the researcher has collaborated on 4 papers that directly or indirectly utilized data and analysis tools produced during the development of TCRpred.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

The TCRpred tool has significantly advanced the field of computational prediction of T cell-epitope interactions in several ways. First, we gathered and curated datasets of T cell-epitope sequence data from different sources as well as in-house generated data.
Better data jointly with new algorithmic developments allowed our tool TCRpred to accurately predict T cell - epitope interactions with higher accuracies with respect to existing tools.
Second, our research demonstrated that computational tools like TCRpred can serve as valuable quality control measures, enhancing the overall quality of both existing and future datasets. Third, TCRpred proved instrumental in providing insights into the specificity of dual α T cells, a subpopulation that constitutes up to 10% of the T cell immune system but has remained poorly characterized until now. Fourth, one of the main promises of T cells interaction predictors is the ability to identify in-silico T cells recognizing specific epitopes directly from T cell repertoire sampled e.g. from blood. Our results indicate that TCRpred offers a robust framework for this in-silico analysis of epitope-specific T cells, particularly when sufficient high-quality training data are available.
Considering the continuously increasing T cell-epitope sequence data, computational tools like TCRpred will become increasingly relevant for the in-silico identification of T cells targeting known viral or cancer epitopes. As these tools continue to improve in accuracy, they hold the potential to facilitate a better understanding of T cell-mediated immunity and simplify the identification or design of candidates for cancer immunotherapy, ultimately reducing time and costs associated with such research endeavors.

msca-1.png