Resurrecting Topiary

[iskandr]

Going to bring this tool back to life with a mission of running (or ingesting predictions from) all the tools which seem to give useful signals of surface presentation of pMHCs.

Predictors

• NetMHCpan 4.1/4.2 (BA/EL, +/- percentile)
• MHCflurry2 (BA/presentation +/- percentile, processing)
• pepsickle, aggregate into a per-peptide score as p_Cterm_cut * (1 - max(p_internal_cut))
• HLApollo
• amino acid properties (net charge, number aromatic residues)

Inputs

In addition to generating mutant peptide sequences from VCF (& RNA expression) files, we should also be able to take the following formats:

• CSV of peptides: peptide predicted across all specified alleles (optional: name, n_flank, c_flank, source_protein)
• CSV of pMHCs: allele, peptide (optional: name, n_flank, c_flank, source_protein)
• CSV of amino acid sequences with mutations specified: aa_before, aa_mut, aa_after(optional: name, source_protein)
• CSV of amino acid sequences without mutations: aa_seq (optional: name, source_protein)
• FASTA of protein sequences without mutations specified (parsed into name and aa_seq)

Outputs

• CSV of all possible pMHC predictions over long sequences or over specified peptides: source_name, allele, peptide, n_flank, c_flank, contains_mutant_residues + columns from all predictors

[bnelsj]

Might be worth adding pvacseq as a supported input type and supporting some inclusion thresholds like percentile.