Background A significant challenge within the field of personalized neoantigen therapies is the determination of which neoantigen targets will elicit durable, therapeutically relevant immune responses. T cell responses can be detected for circa 10–20% of neoepitopes selected for use in vaccines. Screening of memory responses in tumor-infiltrating T cells show much lower rates of 1–2%. Of this small percentage of neoantigens capable of driving an immune response, only a subset will be resistant to methods of tumor immune evasion. Therefore, it is paramount that both these challenges are faced in order to obtain a durable clinical response.

Across different types of neoantigens, the relationship between clonal neoantigens and response to immunotherapy has previously been demonstrated across multiple indications supporting the key role of clonal neoantigens as substrate for T cell recognition of tumors.

Methods Achilles Therapeutics aims to deliver precision immunotherapies specifically targeting clonal neoantigens identified through the Achilles Clonality Engine methodology within our PELEUS^TM bioinformatics platform. The PELEUS^TM platform incorporates a Bayesian approach allowing for the determination of the probability of each potential neoantigen being clonal.

In addition to clonality, and to improve our ability to select for immunogenic neoantigens, we have developed an extensive pipeline for identification of tumor-derived memory T cell responses to clonal neoantigens.

Results Through the use of data obtained by screening circa 10,000 neoantigens for T cell reactivity in expanded tumor-infiltrating lymphocytes, we developed and validated an AI method, NeoRanker, for predicting neoantigen immunogenicity. Using a small set of features incorporating genomic, transcriptomic and proteomic data for training purposes, NeoRanker is able to preferentially enrich our clonal neoantigen list for those capable of driving either CD8+ or CD4+ T cell responses. When benchmarked against well-known tools in the field including BigMHC and Prime, NeoRanker displayed the best performance as measured by the area under the receiver operator characteristic curve.

Conclusions We believe this technology has broad applicability for optimising target selection across all types of personalized neoantigen vaccines and cell therapies.

Trial Registration NCT03997474; NCT04032847; NCT03517917