dt = NULL,
  path = NULL,
  binding_cutoff = 500,
  counts = NULL,
  min_counts = 1,
  assemble = TRUE,
  generate = TRUE,
  predict = TRUE,
  blast = TRUE,
  save = TRUE,
  remove_wt = TRUE



Data table. Input data table from garnish_variants or garnish_jaffa, or a data table in one of these forms:

dt with transcript id:

Column name                 Example input

sample_id                   sample_1
ensembl_transcript_id       ENST00000311936
cDNA_change                 c.718T>A
MHC                         HLA-A*02:01 HLA-A*03:01
                            H-2-Kb H-2-Kb
                            HLA-DRB1*11:07 [second type]

dt with peptide (standard amino-acid one-letter codes only):

Column name                 Example input

sample_id                   <same as above>
pep_wt                      <optional, required for local DAI calculation>
mutant_index                all
                            7 13 14
MHC                         <same as above>

Path to input table (acceptable formats).


Numeric. Maximum consensus MHC-binding affinity that will be passed for IEDB and dissimilarity analysis. Default is 500 (nM). Note: If a peptide binds to any MHC allele in the table below this threshold, IEDB score and dissimilarity will be returned for all rows with that peptide.


Optional. A file path to an RNA count matrix. The first column must contain ENSEMBL transcript ids. All samples in the input table must be present in the count matrix.


Integer. The minimum number of estimated read counts for a transcript to be considered for neoantigen prediction. Default is 1.


Logical. Assemble data table?


Logical. Generate peptides?


Logical. Predict binding affinities?


Logical. Run BLASTp to find wild-type peptide and known IEDB matches?


Logical. Save a copy of garnish_affinity output to the working directory as "ag_output.txt"? Default is TRUE.


Logical. Check all nmers generated against wt peptidome and remove matches? Default is TRUE. If investigating wild-type sequences, set this to FALSE.


A data table of binding predictions including:

  • cDNA_seq: mutant cDNA sequence

  • cDNA_locs: starting index of mutant cDNA

  • cDNA_locl: ending index of mutant cDNA

  • cDNA_type: netMHC prediction tool output

  • frameshift: frameshift variant?

  • coding: wt cDNA sequence

  • coding_mut: mutant cDNA sequence

  • pep_type: type of peptide

  • pep_mut: mutant peptide sequence

  • pep_wt: wt peptide sequence

  • mismatch_s: starting index of mutant peptide sequence

  • mismatch_l: ending index of mutant peptide sequence

  • mutant_index: index of mutant peptide

  • nmer: nmer for prediction

  • nmer_i: index of nmer in sliding window

  • \*_net: netMHC prediction tool output

  • mhcflurry_}*: mhcflurry_ prediction tool output mhcnuggets_}*: mhcnuggets_ prediction tool output DAI: Differential agretopicty index of missense and corresponding wild-type peptide, see garnish_summary for an explanation of DAI. BLAST_A: Ratio of consensus binding affinity of mutant peptide / closest single AA mismatch from blastp results. Returned only if blast = TRUE. clonality info:

    • clone_id: rank of the clone containing the variant (highest equals larger tumor fraction).

    • cl_proportion: The estimated mean tumor fraction containing the clone. If allele fraction and not clonality is used, this is estimated.

    antigen.garnish quality analysis metric results
    • Ensemble_score: average value of MHC binding affinity from all prediction tools that contributed output. 95\

    • iedb_score: R implementation of TCR recognition probability for peptide through summing of alignments in IEDB for corresponding organism. See references herein.

    • IEDB_anno: The best alignment from the IEDB database queried for the sample if applicable.

    • min_DAI: Minimum of value of BLAST_A or DAI values, to provide the most conservative proteome-wide estimate of differential binding between input and wildtype matches.

    • dissimilarity: Calculation from 0 to 1 derived from alignment to the self-proteome, with 1 indicating greater dissimilarity and up to 34-fold odds ratio of immunogenicity (see Citation).

    transcript description:
    • description

    • start_position

    • end_position

    • transcript_end

    • transcript_length

    • transcript_start

    • peptide

    Perform ensemble neoantigen prediction on a data table of missense mutations, insertions, deletions or gene fusions using netMHC, mhcflurry, and mhcnuggets.
    • see list_mhc for compatible MHC allele syntax, you may also use "all_human" or "all_mouse" in the MHC column to use all supported alleles

    • multiple MHC alleles for a single sample_id must be space separated. Murine and human alleles must be in separate rows. See README example.

    • For species specific proteome-wide DAI to be calculated, run human and murine samples separately

    • garnish_score is calculated if allelic fraction or tumor cellular fraction were provided

    Luksza, M, Riaz, N, Makarov, V, Balachandran VP, et al. 2017. A neoepitope fitness model predicts tumour response to checkpoint blockade immunotherapy. Nature. 23;551(7681):512-516 garnish_variantslist_mhcgarnish_summary