827 Streamlining design of safe and effective TCR therapies with AI

BackgroundAdoptive cell therapies with T lymphocytes expressing engineered T cell receptors (TCRs) are one of the most promising approaches to cancer therapy.1 However, the experimentally driven development of novel TCR therapies is limited by the enormous biological variability of peptide:Human Leukocyte Antigen:TCR (pHLA:TCR) complexes. The in silico methods hold the promise to streamline the discovery of novel TCR therapies by reducing costs and time of laboratory research. In particular, the prediction of TCR binding to a target antigen, as well as the prediction of TCR off-target toxicity2 can provide useful insights supporting the development of safe therapies. We aimed at the development of an experimentally validated AI model of pHLA:TCR binding that will help to prioritize and reduce the number of in vitro assays necessary to discover novel TCRs for cancer therapies.MethodsThe limiting factor of successful pHLA:TCR binding modeling is data availability and completeness of TCR characterization. To address this issue, we are building an oncological pHLA:TCR database with paired alpha and beta chain TCR sequences. We are collecting and sequencing tumor and normal samples from 100 cancer patients, as part of an observational clinical trial. Those data are then screened with the Ardigen's ArdImmune Vax platform3 4 to select immunogenic epitopes. T cells that bind those epitopes are subsequently sorted and used to generate TCR sequencing data at single-cell resolution. We use data-driven and simulation-based models to extract insights about the dynamics of a pHLA:TCR system to predict the binding probability and explain the inference made by the model.ResultsWe optimized our data collection pipeline for the cost-efficient acquisition of a large oncological pHLA:TCR dataset. These data will enable us to build efficient models to streamline the development of TCR therapies against cancer.We benchmarked our modeling approach for pHLA:TCR binding against existing solutions5–7 on publicly available data. We also show how focus on model explainability facilitates the detection of model inconsistency of uncertain predictions by expert inspection. Our toxicity assessment solution2 extends the applicability of our system to the prediction of TCR safety profile.ConclusionsThe presented work shows perspectives and limitations of AI-aided TCR therapy development. We present results for our pHLA:TCR binding model, a TCR-toxicity-screening solution, and the study design of our observational clinical trial. Our growing database of pHLA:TCR interactions will enable us to develop highly predictive pHLA:TCR binding models, in particular for oncological targets.AcknowledgementsWe acknowledge funding through the project “Creating an innovative AI-based (Artificial Intelligence) IN SILICO TECHNOLOGY TCRact to launch a NEW SERVICE for designing and optimizing T-cell receptors (TCR) for use in cancer immunotherapies” cofunded by European Regional Development Fund (ERDF) as part of Smart Growth Operational Programme 2014–2020.ReferencesFarkona S, Diamandis EP, Blasutig IM. Cancer immunotherapy: the beginning of the end of cancer? BMC Med 2016;14:73. PMCID: PMC4858828.Murcia Pienkowski VA, Mazzocco G, Niemiec I, Sanecka-Duin A, Krol P, Myronov O, Skoczylas P, Kaczmarczyk J, Blum A. Off-target toxicity prediction in cellular cancer immunotherapies [Internet]. Cytotherapy. 2021;S96. Available from: http://dx.doi.org/10.1016/s1465324921004229.Stepniak P, Mazzocco G, Myronov A, Niemiec I, Gruba K, Skoczylas P, Sanecka-Duin A, Drwal M, Kaczmarczyk J. AI-augmented design of effective therapeutic cancer vaccines and adoptive cell therapies. Journal For Immunotherapy Of Cancer. Bmc Campus, 4 Crinan St, London N1 9xw, England; 2019.Mazzocco G, Niemiec I, Myronov A, Skoczylas P, Kaczmarczyk J, Sanecka-Duin A, Gruba K, Król P, Drwal M, Szczepanik M, Pyrc K, Stȩpniak P. AI aided design of epitope-based vaccine for the induction of cellular immune responses against SARS-CoV-2. Front Genet. 2021;12:602196. PMCID: PMC8027494.Weber A, Born J, Rodriguez Martínez M. TITAN: T-cell receptor specificity prediction with bimodal attention networks. Bioinformatics. 2021;37(Suppl_1):i237–i244. PMCID: PMC8275323.Springer I, Besser H, Tickotsky-Moskovitz N, Dvorkin S, Louzoun Y. Prediction of specific TCR-peptide binding from large dictionaries of TCR-Peptide Pairs. Front Immunol 2020;11:1803. PMCID: PMC7477042.Jurtz VI, Jessen LE, Bentzen AK, Jespersen MC, Mahajan S, Vita R, Jensen KK, Marcatili P, Hadrup SR, Peters B, Nielsen M. NetTCR: sequence-based prediction of TCR binding to peptide-MHC complexes using convolutional neural networks [Internet]. Available from: http://dx.doi.org/10.1101/433706