PGV-001 (Personalized Genomic Vaccine) is a Phase I clinical trial at Mount Sinai testing the safety, feasibility, and immunogenicity of a neoantigen + Poly-ICLC vaccine derived from exome and RNA sequencing [1]. The personalized component of the vaccine consists of ten peptides (25 amino acids in length) predicted to contain highly expressed mutated cytotoxic T-cell epitopes. Selecting peptides for each patient requires a number of standard bioinformatics tasks such as somatic variant calling, sequence-based HLA typing, and class I peptide-MHC binding prediction, as well as new methods to determine the coding sequence for each identified mutation, quantify the RNA support for each variant allele, and scoring each vaccine peptide. We present the overall design of PGV, its implementation as an open source computational pipeline, and several independently useful libraries that implement aspects of mutant epitope prediction.

One of the unique characteristics of the PGV pipeline is that the amino acid sequence of a candidate vaccine peptide is determined from RNA reads overlapping a variant [2]. Consequently, variants which lack any evidence in RNA cannot be used in PGV. Determining a coding sequence from RNA allows us to capture atypical splicing around a mutation as well as phase somatic variants with nearby germline variants. Once a mutant coding sequence is established, candidate vaccine peptides are ranked by combining their predicted Class I MHC binding affinity with allele-specific expression [3]. We demonstrate the impact of different ranking criteria on vaccine efficacy using a tumor mouse model.

Execution and dependency management of the PGV pipeline is orchestrated using a a novel workflow engine [4, 5] which allows us to run complex tools across many machines with fault tolerance and reproducibility. The pipeline can be run on single machines, local clusters, or on public infrastructure using the Google Cloud Platform. We hope that by making the PGV pipeline and its underlying components available as open source software we can facilitate faster experimentation in the design of personalized cancer vaccines.






Citation Format: Alexander Rubinsteyn, Isaac Hodes, Timothy O'Donnell, Sebastien F. Mondet, John P. Finnigan, Philip Friedlander, Rachel Sabado, Nina Bhardwaj, Jeffrey Hammerbacher. Computational pipeline for a personalized genomic vaccine trial [abstract]. In: Proceedings of the Second CRI-CIMT-EATI-AACR International Cancer Immunotherapy Conference: Translating Science into Survival; 2016 Sept 25-28; New York, NY. Philadelphia (PA): AACR; Cancer Immunol Res 2016;4(11 Suppl):Abstract nr A022.