Neoantigen prediction

The neoantigen prediction pipeline 'the myNEO ImmunoEngine' encloses the industry-standard wireframe for neoantigen prediction but is enhanced with a variety of filters/ranking mechanisms based on biological processes to retain the neoantigens providing the most clinical benefit. It incorporates advanced features that enable prediction of novel neoantigen types in low mutational burden malignancies and limit the number of false positives leading to improved vaccine efficacy.

Fuelled by advances in genomic and proteomic technologies, personalised oncology promises to innovate cancer therapy and to target previously untreatable tumours. Moreover, personalised cancer immunotherapies offer the promise of low toxicity and high specificity and the opportunity to treat human malignancies resistant to current therapies. A critical part of developing such strategies is finding those antigens that have the potential to elicit a sufficiently strong immune response, so-called neoantigens. For the past years, myNEO has been developing the neoantigen prediction platform, providing a robust pipeline focused on the identification and prediction of neoantigens.

Graphical representation of the basis of the myNEO ImmunoEngine wireframe

The process starts with the identification of all tumour somatic non-synonymous mutations (NSM). To accommodate this discovery, Whole-Genome Sequencing (WGS) or Whole-Exome Sequencing (WES) data from matched tumour and normal DNA is required for each patient. Following alignment of these reads to the human reference genome, somatic genetic alterations in the tumour genome are detected using variant-calling algorithms.

To avoid the incorrect classification of germline variants as neoantigens and to guarantee the expression of the identified variants, both RNA and DNA from tumour tissue is used by the most up-to-date software packages to compare with DNA from the matched normal tissue (Sahin & Türeci, 2018; Xu, 2018). Most variant calling algorithms have already proven to reliably call single nucleotide variants (SNVs) since they are the most abundant type of tumour mutation (Vogelstein et al. 2013) and because of the relative simplicity and reliability of identifying sequence changes of one base pair (Turajlic et al. 2017). The myNEO ImmunoEngine is focused on specificity, and as such, uses a two-pass variant calling method, wherein two variant callers using different methods are confronted with the dataset, and only variants passing all customised filtering are selected for further neoantigen prediction.

"The expansions make the myNEO ImmunoEngine one of the most advanced available neoantigen prediction pipelines which allow for higher-confident neoantigen predictions whilst lowering false-positive rates"

Besides SNVs, indels and gene fusions can lead to highly immunogenic frameshifts (Turajlic et al. 2017; Türeci et al. 2016; Yang et al. 2010), albeit their detection remains challenging (Sun et al. 2016). Moreover, alternative splicing events (neoisoforms) and retrotransposable elements can also lead to new epitopes and are therefore taken into consideration in the analysis.

The somatic variant calling step in the ImmunoEngine renders a set of high confidence somatic variants that can be used for the neoantigen prediction and prioritisation, i.e. selection of the variant-derived neoantigens that are the most likely to be presented on HLA alleles and thus successfully passing all antigen processing steps. Most companies focus purely on MHC-I, however there can be synergies between MHC-II-mediated CD4-positive T-cell activation and MHC-I-mediated CD8 T-cell activation. The myNEO ImmunoEngine delivers analysis and prioritisation of both MHC-I and MHC- II potential neoepitopes.

Overview of the myNEO ImmunoEngine

Prioritisation is further achieved with the following set of filters:

  • Level of expression of the peptide as intuited from the RNA-Seq transcript-level analysis; neopeptides derived from highly expressed mRNAs are in general more likely to be translated and thus picked up by the MHC presentation machinery.
  • Allelic frequency of the variant responsible for the neoepitope: variants with higher allele frequency are present through more of the tumour and thus more interesting for a therapeutic approach.
  • Dissimilarity to self: neoepitopes have to be not only different to the original peptides they are derived from due to mutation, but also from any other possible antigen from the native proteome, as the identity to such antigens implies that the neoepitope triggers self-tolerance (Yarchoan et al. 2017).
  • Overall coverage and read depth of the genomic feature responsible for the neopeptide: a feature conclusively identified with many supporting sequencing reads lends more confidence to the neoepitopes it ultimately produces.
  • Impact of the somatic variant: neoantigens originating from a somatic variant with a high impact on tumour progression are more interesting to target because the chance of resistance is reduced.
  • Assessment of neopeptide immunogenicity is obtained through dissimilarity to self (see above) and through similarity to a dataset of known immunogenic peptides to ensure the selection of the candidates with the highest immunogenic potential.

The neopeptide’s ability to be processed, transported up to the MHC molecule and stably bind to its components is also assessed through several prediction tools taking advantage of novel machine learning algorithms to further select the highest confidence set. In addition, myNEO has developed a deep learning model that improves the prediction of neoantigens successfully undergoing the whole antigen presentation pathway. This model is based on patterns detected in MS peptide-HLA ligandome data, conveniently named the neoMS algorithm. By training a neural network on MS data representative of the complete antigen presentation process (including peptide processing, peptide transportation and localization, and MHC binding), it predicts a more complete process than its counterparts using biochemical data based on just one of these steps (mostly the MHC-binding affinity).

Important steps in antigen processing and presenting are not considered yet by most prediction pipelines, be it their handling by various molecular machinery, their potential immunogenicity, etc. Analysing new ranking methodologies based on these parameters help refine the prioritisation step of myNEO's prediction pipeline, thus ensuring that only the highest confidence neoantigens are selected. myNEO is currently involved in several innovation projects where new tools are geared towards addressing these issues and are being evaluated in order to achieve optimal detection of highly efficient neoantigens.

Want to learn more about this?

Contact us!

Vogelstein, B., Papadopoulos, N., Velculescu, V. E., Zhou, S., Diaz, L. A., & Kinzler, K. W. (2013). Cancer genome landscapes. Science (New York, N.Y.), 339(6127), 1546–1558. https://doi.org/10.1126/science.1235122
Sahin, U., & Türeci, Ö. (2018). Personalized vaccines for cancer immunotherapy. Science (New York, N.Y.), 359(6382), 1355–1360. https://doi.org/10.1126/science.aar7112
Xu, C. (2018). A review of somatic single nucleotide variant calling algorithms for next-generation sequencing data. Computational and Structural Biotechnology Journal, 16, 15–24. https://doi.org/10.1016/J.CSBJ.2018.01.003
Turajlic, S., Litchfield, K., Xu, H., Rosenthal, R., McGranahan, N., Reading, J. L., ... Swanton, C. (2017). Insertion-and-deletion-derived tumour- specific neoantigens and the immunogenic phenotype: a pan-cancer analysis. The Lancet Oncology, 18(8), 1009–1021. https://doi.org/10.1016/S1470-2045(17)30516-8
Türeci, Ö., Vormehr, M., Diken, M., Kreiter, S., Huber, C., & Sahin, U. (2016). Targeting the Heterogeneity of Cancer with Individualized Neoepitope Vaccines. Clinical Cancer Research: An Official Journal of the American Association for Cancer Research, 22(8), 1885–1896. https://doi.org/10.1158/1078-0432.CCR-15-1509
Yang, H., Zhong, Y., Peng, C., Chen, J.-Q., & Tian, D. (2010). Important role of indels in somatic mutations of human cancer genes. BMC Medical Genetics, 11, 128. https://doi.org/10.1186/1471-2350-11-128
Sun, Z., Bhagwate, A., Prodduturi, N., Yang, P., & Kocher, J.-P. A. (2016). Indel detection from RNA-seq data: tool evaluation and strategies for accurate detection of actionable mutations. Briefings in Bioinformatics, 18(6), bbw069. https://doi.org/10.1093/bib/bbw069
Yarchoan, M., Johnson, B. A., Lutz, E. R., Laheru, D. A., & Jaffee, E. M. (2017). Targeting neoantigens to augment antitumour immunity. Nature Reviews. Cancer, 17(4), 209–222. https://doi.org/10.1038/nrc.2016.154