pVACseq logo


SNV and Indel support

pVACseq offers epitope binding predictions for missense, in-frame insertion, in-frame deletion, protein-altering, and frameshift mutations.

VCF support

pVACseq uses a VCF file as its input. This VCF file must contain sample genotype information and be annotated with the Ensembl Variant Effect Predictor (VEP). See the Input File Preparation section for more information.

No local install of epitope prediction software needed

pVACseq utilizes the IEDB RESTful web interface. This means that none of the underlying prediction software, like NetMHC, needs to be installed locally.


We only recommend using the RESTful API for small requests. If you use the RESTful API to process large VCFs or to make predictions for many alleles, epitope lengths, or prediction algorithms, you might overload their system. This can result in the blacklisting of your IP address by IEDB, causing 403 errors when trying to use the RESTful API. In that case please open a ticket with IEDB support to have your IP address removed from the IEDB blacklist.

Support for local installation of the IEDB Analysis Resources

pVACseq provides the option of using a local installation of the IEDB MHC class I and class II binding prediction tools.


Using a local IEDB installation is strongly recommended for larger datasets or when the making predictions for many alleles, epitope lengths, or prediction algorithms. More information on how to install IEDB locally can be found on the Installation page (note: the pvactools docker image now contains IEDB).

MHC Class I and Class II predictions

Both MHC Class I and Class II predictions are supported. Simply choose the desired prediction algorithms and HLA alleles during processing and Class I and Class II prediction results will be written to their own respective subdirectories in your output directory.

By using the IEDB RESTful web interface, pVACseq leverages their extensive support of different prediction algorithms.

In addition to IEDB-supported prediction algorithms, we’ve also added support for MHCflurry and MHCnuggets.

MHC Class I Prediction Algorithm
















MHC Class II Prediction Algorithm









Comprehensive filtering

Automatic filtering on the binding affinity ic50 (nm) value narrows down the results to only include “good” candidate peptides. The binding filter threshold can be adjusted by the user for each pVACseq run. pVACseq also support the option of filtering on allele-specific binding thresholds as recommended by IEDB. Additional filtering on the binding affitinity can be manually done by the user by running the standalone binding filter on the filtered result file to narrow down the candidate epitopes even further or on the unfiltered all_epitopes file to apply different cutoffs.

Readcount and expression data are extracted from an annotated VCF to automatically filter with adjustable thresholds on depth, VAF, and/or expression values. The user can also manually run the standalone coverage filter to further narrow down their results from the filtered output file.

If the input VCF is annotated with Ensembl transcript support levels (TSLs), pVACseq will filter on the transcript support level to only keep high-confidence transcripts of level 1. This filter can also be run standalone.

As a last filtering step, pVACseq applies the top score filter to only keep the top scoring epitope for each variant. As with all previous filters, this filter can also be run standalone.

Incorporation of proximal germline and somatic variants

To incorporate proximal variants into the neoepitope predictions, users can provide a phased VCF of proximal variants as an input to their pVACseq runs. This VCF is then used to incorporate amino acid changes of nearby variants that are in-phase with a somatic variant of interest. This results in corrected mutant and wildtype protein sequences that account for proximal variants when MHC binding predictions are performed.

NetChop and NetMHCstab integration

Cleavage position predictions are added with optional processing through NetChop.

Stability predictions can be added if desired by the user. These predictions are obtained via NetMHCstabpan.

Reference proteome similarity analysis

This optional feature will search for an epitope in the reference proteome using BLAST to determine if the epitope occurs elsewhere in the proteome and is, therefore, not tumor-specific.