Create a tool that aggregates and prioritizes/filters fusion neoantigens

[malachig]

Objectives of the tool:

• Create a subdirectory to store the essential information (files) needed for review
• Create a Tiered and Annotated report table that makes it more clear which fusion candidates are worthy of detailed review.

Minimal Files needed to perform review:

• (FI web report) rnaseq/fusioninspector_evidence/finspector.fusion_inspector_web.html
• (FI coding effect TSV) finspector.FusionInspector.fusions.abridged.tsv.annotated.coding_effect
• (pVACfuse aggregated epitopes) $SAMPLE-tumor-exome.all_epitopes.aggregated

Steps

1. Create copy of all files above in a new subdirectory

2. Create a prioritized candidate list based on the FI coding effect TSV file. Filtering criteria. Must meet all of the following criteria:
   Read Support
   (a) Junction + Spanning counts > 5; (b) junction >= 1;

NOT a readthrough. Defined as:
[Left Chr and Right Chr are different] OR
[chromosome are the same BUT Left Strand and Right Strand are different] OR
[chromosome and strand are the same BUT ABS(Left Pos - Right Pos) < 1,000,000] OR
[Fusion GeneA Name OR Fusion GeneB Name matches a known fusion driver gene]

Anchor support (?)
Require ​​LargeAnchorSupport == YES

3. Eliminate candidates that do not give rise to neoantigens at all
   Calculate a count of unique peptides (“Best Peptide”) from the pVACfuse aggregated epitopes that match the GeneA_GeneB pairing (e.g. “KANK4_ALK”). If this count is 0, the candidate fusion event will not be marked in “REVIEW” Tier, otherwise assigned to “POOR” Tier.

4. Extract 51-mer peptide sequence centered on gene fusion junction.
   This was being done manually by use of BLAT and string match for junction peptide sequences. Figure out a way to do this in a more automated fashion with info from files from: FusionInspector, StarFusion, and pVACfuse.

5. Create the final review table file (analogous to pVACview main table).

• All/Most of the columns from FusionInspector
• FFPM value
• Full fusion amino acid sequence (with GeneA/GeneB junction marked somehow)
• Unique neoantigen peptide count (total, and per HLA allele). Where multiple transcript pairs give rise to the same exact peptides, define a transcript-pair set, and pick a representative transcript pair.
• Other info from pVAFfuse: IC50 MT, %ile MT, Expr, Read Support

One time tasks

Define a reference list of known fusion driver genes
Cancer Gene Census entries where “Role In Cancer” includes fusion. Could be refined to consider tumor type. Or manually curated to a higher confidence list.

Abbreviations:
FI = FusionInspector

[susannasiebert]

Reading over this list I'm wondering if some of these steps should happen in pvacfuse itself. In my mind, the pVACfuse aggregated report is what should be used for protizing fusion neoantigens, so if the format and available fields don't match what is needed, we should update pVACfuse to create a better aggregated report instead of using the aggregated report + some other inputs to create yet another report. This would be especially important since we would eventually like pVACview to support output files from pVACfuse so that candidates can be evaluated in the same interface.

A few items that stick out to me:

• We already allow a starfusion input to be supplied to pVACfuse in order to obtain filter on read support (junction + spanning). Read support is being used to tier the candidates in the aggregate report. Do we need to change the default cutoff for read support and and also include the junction counts themselves with a new cutoff?
• For the final review file, I think we should consider to update pVACfuse and the fields it includes in the aggregated report instead of creating a new file. This would require us to feed FI output into pVACfuse but that wouldn't be difficult.
• For the 51 mer, is there are reason to not use the pvacfuse generate_protein_fasta command?
• Candidates from read through fusions and without LargeAnchorSupport could get their own aggregate report tiers.