7.0.0

What's Changed

This is a major version release. Please note that pVACtools 7.0 is not guaranteed to be backwards-compatible and certain changes could break old workflows.

New Tools

pVACseq now has the option of running machine learning (ML)-based neoantigen prioritization predictions. The ML predictor uses a trained random forest model to predict whether neoantigen candidates in the aggregated report should be evaluated as “Accept”, “Reject”, or “Pending” based on a comprehensive set of features derived from binding affinity predictions, expression data, and variant characteristics. The ML predictor can be enabled by adding the --run-ml-predictions parameter to a pVACseq run. More details can be found in the output file documentation, the vignette.

New Features

• pVACtools now supports multiple additional prediction algorithms:
  • MixMHCpred (class I binding score and percentile)
  • MixMHC2pred (class II binding score and percentile)
  • PRIME (class I immunogenicity score and percentile)
  • ImmuScope (class II immunogenicity score)
• In order to support a more comprehensive evaluation of candidates, aggregate binding, presentation, and immunogenicity information is now available in the final reports and is used to filter, prioritize, and tier candidates:
  • Best and median binding percentiles, presentation percentiles and immunogenicity percentiles are now calculate in the all_epitopes.tsv, filtered.tsv, and aggregated.tsv files in addition to the previously available combined percentiles that were aggregating percentile ranks over all prediction algorithms regardless of algorithm type.
  • Three new parameters - --binding-percentile-threshold, --presentation-percentile-threshold, and --immunogenicity-percentile-threshold replace the old --percentile-threshold. These three new thresholds have been updated to use a default of 2.0 instead of not having a default. This means that filtering and tiering will now by default include evaluation of binding, presentation, and immunogenicity percentiles.
  • The aggregate report PoorBinding tier now evaluates the IC50 binding affinity as well as the binding percentile. Candidates failing either threshold (when a conservative --percentile-threshold strategy is selected, default) or both thresholds (when a exploratory --percentile-threshold-strategy is selected) will be binned in this tier when all other evaluation criteria are passed.
  • Two new tiers, PoorPresentation and PoorImmunogenicity, are added to bin candidates that failed the --presentation-percentile-threshold or --immunogenicity-percentile--threshold, respectively, when all other evaluation criteria are passed.
  • pVACvector has been updated to work on the binding percentile instead of the combined percentile. The corresponding parameter has been renamed to --binding-percentile-threshold with a new default of 2.0. The junction output file header recording each junction’s binding percentile has been updated from percentile to binding_percentile to reflect this change.
  • pVACview has been updated to display more information regarding immunogenicity and presentation scores.
  • pVACtools runs that do not use a binding predictor would previously skip the binding filter, top score filter, and the aggregate report creation. These steps will now be run.
• The --top-score-metric2 has been updated for sorting candidates and determining the criteria for selecting the Best Peptide (in the aggregate report) and top candidate (in the top score filter). It is now a list of criteria to consider. All listed criteria are assigned a rank and the sum of those ranks is used. By default both the IC50 (ic50) and the combined percentile (combined_percentile) are used. Other allowed values are the binding percentile (binding_percentile), the presentation percentile (presentation_percentile), and the immunogenicity percentile (immunogenicity_percentile). Any number and combination of these five criteria may be specified.
• Not all prediction algorithms supported by pVACtools may support a percentile rank. In order to alleviate this issue, and to provide percentile ranks that have been consistently calculated, we have run predictions for all class I algorithms supported by pVACtools on 100,000 reference peptides each in lengths 8-11 and for the most common 1,000 human class I MHC alleles. These predictions support a new feature in pVACtools: normalized percentiles (--use-normalized-percentiles). With this option enabled, any of the pVACtools pipelines will calculate normalized percentiles scores for all predicted neoantigen candidates and selected prediction algorithms. These normalized percentile ranks will be used in place of percentile ranks calculated by the algorithms natively. Predictions for allele or lengths we have not calculated reference scores will result in NA percentile ranks. Turning on this option in class II runs or with non-human data will be ignored. The peptides used in our predictions and the raw scores we calculated are available at https://github.com/griffithlab/pvactools_percentiles_data.
• In pVACbind and pVACfuse the Mutation column name in the various report files has been renamed to Index in order to ensure consistency between these and other tools.

New Contributors

• @jyao36 made their first contribution in #1333

Full Changelog: v6.1.1...v7.0.0