OpenVax is a collection of open-source Python libraries and command-line tools for finding the mutations in a patient's tumor that T cells could recognize, and choosing which of them to put in a vaccine. Vaccines designed with these tools have been given to patients in clinical trials.
The libraries are maintained at the Personalized Immunotherapy Research Lab (PIRL) at UNC Chapel Hill.
Each step below is a separate package that can be used on its own.
Chooses the mutant peptides for a personalized vaccine. Scores each candidate by predicted MHC presentation and by how much tumor RNA supports it, then outputs ranked peptides, mRNA constructs or reports for clinical review. Also accepts results from LENS or pVACseq.
pip install vaxrank
Determines the mutant protein sequence the tumor expresses by assembling the RNA reads that cover each mutation, so nearby germline or somatic variants and splicing changes come from the reads rather than the reference genome.
pip install isovar
Predicts what DNA variants do to genes, transcripts and proteins: substitutions, frameshifts, stop codons, splice and structural variants. Reads VCF and MAF files and returns the mutant protein where it can be determined.
pip install varcode
One Python interface to many MHC binding, presentation, processing and immunogenicity predictors, including NetMHCpan, MHCflurry, MixMHCpred and BigMHC. Switching or comparing predictors is a one-line change.
pip install mhctools
Predicts which peptides MHC class I molecules present, with pretrained models for binding affinity, antigen processing and presentation. Runs from the command line or Python and can be retrained on new data.
pip install mhcflurry
DocumentationSourcePaper (2018)MHCflurry 2.0 paper (2020)Book chapter (2020)
Predicts, filters and ranks MHC-presented peptides from any source, including tumor mutations and viral proteins. Built on mhctools.
pip install topiary
The end-to-end workflow used in the Mount Sinai trials: aligns tumor and normal sequencing reads, calls somatic variants and runs Vaxrank. Packaged with Snakemake and Docker.
Local access to Ensembl genome annotation (genes, transcripts, exons and their sequences) for human and many other species.
pip install pyensembl
Fast parsing of GTF gene annotation files into data frames.
pip install gtfparse
Downloads, checks and caches the reference data that PyEnsembl and other packages use.
pip install datacache
varlens (command-line tools for variants and reads), pepdata (IEDB data and amino acid properties), mhcgnomes (parses MHC allele names), sercol and serializable (shared base classes).
The neoantigen-vaccine-pipeline runs everything from raw sequencing reads to ranked vaccine peptides inside one Docker image. You need tumor and normal whole-exome sequencing and tumor RNA-seq as gzipped FASTQ files, the patient's MHC class I alleles, and a machine with at least 16 cores. No cluster is required.
docker pull openvax/neoantigen-vaccine-pipeline:latest
docker run openvax/neoantigen-vaccine-pipeline:latest -hProcessed b37decoy, GRCh38 and mm10 references are available in Google Cloud. For GRCh38:
gsutil -m cp gs://reference-genomes/grch38.tar.gz /path/to/reference-genome/
cd /path/to/reference-genome && tar -zxvf grch38.tar.gz
chmod -R a+w grch38They are mounted as Docker volumes: /inputs holds the FASTQ files and a config file, /outputs receives results, and /reference-genome holds the reference data. All three must be world-writable, because the pipeline runs as an unprivileged user.
A YAML file in /inputs lists the sample's files, up to six MHC class I alleles (each must be supported by NetMHCpan) and tool settings. Paths are relative to the mounted directories. Start from the example config. Paired-end reads use r1 and r2 entries with type: paired-end; data from several lanes goes in separate fragments with distinct fragment_id values.
The test data is a small set of reads covering the IDH1 R132H mutation, with the tumor DNA reads reused as RNA.
cd /path/to/inputs
URL=https://github.com/openvax/neoantigen-vaccine-pipeline/raw/master
wget $URL/test/idh1_config_grch38.yaml
wget $URL/datagen/idh1_r132h_normal.fastq.gz
wget $URL/datagen/idh1_r132h_tumor.fastq.gz
docker run -it \
-v /path/to/inputs:/inputs \
-v /path/to/outputs:/outputs \
-v /path/to/reference-genome:/reference-genome \
openvax/neoantigen-vaccine-pipeline:latest \
--configfile=/inputs/idh1_config_grch38.yamlThe first run takes a few extra minutes while files are downloaded and cached. Results include ranked variants and vaccine peptides as text and PDF reports; the test should report a single IDH1 R132H variant.
To call somatic variants only, leave the tumor RNA and HLA alleles out of the config and the pipeline writes Mutect and Strelka VCFs instead (example config). Intermediate outputs and other options are described in the pipeline README.
Adapted from a 2020 post by Julia Kodysh.
Four trials at Mount Sinai have tested PGV-001, a personalized vaccine of up to 10 synthetic long peptides per patient, chosen by the OpenVax pipeline and given with the adjuvant poly-ICLC.
Adjuvant PGV-001 after standard treatment of head and neck, lung, breast or bladder cancer, or multiple myeloma. The vaccine was safe and induced T-cell responses in every vaccinated patient.
PGV-001 with tumor treating fields, added to standard treatment (surgery, radiation and temozolomide) for newly diagnosed glioblastoma. Enrollment is complete; follow-up continues.
PGV-001 with the checkpoint inhibitor atezolizumab for advanced urothelial cancer. The combination was safe and induced T-cell responses in all patients.
Adjuvant PGV-001, alone or with the immune stimulant CDX-301 (FLT3 ligand). Enrollment is complete (27 patients); results are pending.
A dendritic cell vaccine loaded with each patient's neoantigens, given with nivolumab after surgery for hepatocellular carcinoma or colorectal cancer metastatic to the liver. Neoantigens are selected with the OpenVax pipeline.
OpenVax started in Hammer Lab, Jeff Hammerbacher's lab at Mount Sinai, where the first of these libraries were written. The group then built the OpenVax pipeline to select vaccine targets for Nina Bhardwaj's Vaccine and Cell Therapy Lab, and for several years was its own small research group within Mount Sinai.
The group has since wound down. Mount Sinai still uses the software to design vaccines, and development continues at PIRL at UNC Chapel Hill.
Wrote most of the OpenVax libraries and still maintains them. Faculty in Genetics and Computational Medicine at UNC Chapel Hill and a member of the UNC Lineberger Comprehensive Cancer Center. Co-leads the Personalized Immunotherapy Research Lab.
Built the vaccine pipeline used in the Mount Sinai trials and still works on it. Works at Freenome.
Created MHCflurry and still contributes to it. Now at Open Athena, a nonprofit that helps academic labs build scientific AI models.
Directs the Bioinformatics Program at the Mount Sinai Tisch Cancer Center and uses OpenVax to design vaccines there.
Founded Hammer Lab, where OpenVax began. Now founder and CEO of Open Athena.
Worked on early versions of PyEnsembl, Varcode, mhctools and Topiary. Now runs Sequence Software and Purple Computer.
Worked on the first version of the PGV trial pipeline. Now at Open Athena.
Other contributors include Arun Ahuja, Arman Aksoy, Seb Mondet and Walid Ahmad.