Neolution-live

Pipeline for neo-antigen prediction

Introduction

This pipeline performs neo-antigen predictions using netMHCpan and netChop. RNA expression data and 'similarity-to-self' filters can additionally be used to further increase the precision of the predictions. Alternatively, we trained a random forest classifier on mass spectrometry data, which integrates predicted affinity rank, proteasomal processing scores and RNA expression values to a combined model score. This score lies in the range [0-1] and gives the probability of peptide presentation and has been shown to substantially increase the precision and sensitivity over the conventional use of binary cutoff values for the various predicted parameters. This is now our preferred prediction method.
Suggested model score cutoff values are:

• 0.01 for TIL screens (more inclusive; we picked up low magnitude TIL hits at low probability scores)
• 0.02 for PBMC screens (more stringent; unlikely to pick up low magnitude responses with low prob scores in peripheral blood)

The pipeline can be supplied with different forms of input. Either As input, the pipeline expects a tsv file with affected germline and tumor transcripts. Additional variant/transcript information can be provided and will be transferred into the output. See Input file format paragraph for more information regarding input file generation.

Minimal usage example

The only required argument is the input file, other settings will will take default values if these are not supplied

neolution --mhc A0201 --model 0.02 --length 9 /path/to/variants.tsv

The call will start neo-antigen predictions for variants.tsv, HLA-A*02:01, 9-mer peptides, applying a model prediction cutoff score of 0.02.

neolution /path/to/variants.tsv

Run Neolution using default settings (see below)

Commandline arguments

Required

1. full input file path

Optional

1. netChop processing cutoff
2. RNA expression cutoff
3. single sequence input (fasta input: not paired tumor-normal, no rna expression)
4. structural variant predictions
5. simple self-similarity check (9-, 10-, 11-mers)
6. extended self-similarity check (9-mers only)
7. use self-epitope list
8. use database for peptide affinity lookups (9-mers, netMHCpan-2.4 only)
9. netMHCpan version

NOTE: self-similarity checking requires predicted self-epitope lists of matching HLA & peptide length

Input file format

Required input format is a wide, tab-separated table with the following columns. Neolution-prep generates all required data and provides required input file.

Variant

variant_id	chromosome	start_position	end_position	variant_strand	ref_allele	alt_allele

Gene/Transcript

gene_id	transcript_id	transcript_strand	hugo_symbol	variant_classification	transcript_remark	transcript_extension	nmd_status	nmd_remark

DNA/RNA

dna_ref_read_count	dna_alt_read_count	dna_total_read_count	dna_vaf	rna_ref_read_count	rna_alt_read_count	rna_total_read_count	rna_vaf	rna_alt_expression	rna_expression

Effect

codon_germline	codon_tumor	aa_germline	aa_tumor	aa_pos_germline	aa_pos_tumor_start	aa_pos_tumor_stop	peptidecontextnormal	peptidecontexttumor

---

Rscript performPredictions.R --help

Usage: performPredictions.R [OPTIONS]

OPTIONS
-f FILE, --file=FILE
Full path to file containing variant calls (required)

-m MHC, --mhc=MHC
MHC/HLA allele, formatted as follows: A0201 (required)

-l LENGTH, --length=LENGTH
Peptide length (required)

-d MODEL, --model=MODEL
Random forest model score cutoff (optional, suggested cutoff 0.02 for PBMC, 0.01 for TIL screens)

-r RANK, --rank=RANK
netMHCpan rank cutoff (optional, suggested is 2.8 ~500nM for A*02:01)

-a AFFINITY, --affinity=AFFINITY
netMHCpan affinity cutoff (optional)

-p PROCESSING, --processing=PROCESSING
netChop processing score cutoff (optional, default: >= 0.5)

-e EXPRESSION, --expression=EXPRESSION
RNA expression cutoff (optional, default: > 0; use -1 for no filtering)

--single
Single sequence predictions (not paired normal-tumor) (optional, default: FALSE)

--structural
Structural variant predictions (optional, default: FALSE)

--selfsim
Perform simple self-similarity check; compatible with 9-, 10-, 11-mers (optional, default: FALSE)

--extselfsim
Perform extended self-similarity check; only compatible with 9-mers (optional, default: FALSE)

--selflist
Add predicted self-epitopes to self-similarity check, requires length- & HLA-matched selflist (optional, default: FALSE)

--fasdb
Look up peptide affinity in FASdb, predict if not found; only compatible with 9-mers & netMHCpan-2.4 (optional, default: FALSE)

--panversion
Use different version of netMHCpan; must be installed in path specified in runConfig.R (optional, default: 4.0)

--verbose
Be chatty (optional, default: FALSE)

-h, --help
Show this help message and exit