RIFTT

Robust Identification of Fusion genes in Tumor Transcriptomes.

RIFTT calls fusion genes from RNA-seq data with Arriba and FusionCatcher and applies knowledge-based filtering and three scoring metrics (Promiscuity Score, Fusion Transcript Score, Robustness Score) to prioritise robust candidates. It is bundled in a Singularity container and runs in two parts: detection per sample (Part 1) and cohort-level filtering (Part 2).

Requirements

• Singularity or Apptainer, version 3.6 or newer

• at least 40 GB RAM

• Reference files:

  • genome FASTA and gene annotation GTF (GENCODE, https://www.gencodegenes.org)
  • STAR index built without --sjdbGTFfile / --sjdbOverhang
  • FusionCatcher genomic database (https://sourceforge.net/projects/fusioncatcher/files/data)

Build the container

git clone https://github.com/pkerbs/RIFTT.git
cd RIFTT
sudo singularity build RIFTT.sif RIFTT.def       # or: singularity build --fakeroot

RIFTT.sif is expected next to the helper scripts (or set RIFTT_SIF).

Configure

cp config/params.conf.example config/params.conf

Edit config/params.conf (paths, genome build, threads, ...). Both helper scripts read this file. All options are described in config/params.conf.example.

Run

Part 1 - detection (once per sample):

./RIFTT_part1.sh <sample_name> <fastq_folder>

FASTQ files are matched as <sample_name>*R1.fastq.gz / *R2.fastq.gz.

Part 2 - filtering (once, over the whole cohort):

./RIFTT_part2.sh

Clinical table (Part 2 input)

An .xlsx file with one row per sample and four columns:

column	required	content
cohort	yes	cohort / group label; PS, maxPS and RS are computed over all samples of the run, and the plots are drawn per cohort
sample	yes	sample name, identical to the one used in Part 1
Karyotype	no	ISCN karyotype string, e.g.46,XX,t(6;9)(p23;q34)[12]; leave empty if unknown
otherCyto	no	fusions known for this sample from other diagnostics; leave empty if none

Karyotype and otherCyto are annotation only - they fill the karyo / mol columns of the output and the oncoprint, and do not affect filtering or ev_level. A clinical table with only cohort and sample filled in is valid.

otherCyto holds comma-separated entries, each one of:

entry	meaning	setsmol
GENE1::GENE2	fusion confirmed by other diagnostics	Y
GENE#	one partner screened / suspected	S
GENE1!GENE2	fusion explicitly tested and found negative	N

Example values: PML::RARA · RUNX1::RUNX1T1,CDC2L1::SLC35E2 · DEK!NUP214

Output

Part 2 writes to <OUTPUTFOLDER>/filter_results/run_<timestamp>/:

file	content
resultTable.xlsx	all evaluated fusion calls with their metrics
filterrun.RData	the R workspace of the run
plots/Violinplot_PS.png, Violinplot_FTS.png	PS / FTS distributions, known vs unknown
plots/TPM-FTS_3D_plot.html	interactive TPM vs FTS scatter
plots/Oncoprint_Karyo_MDx_RNAseq.png	oncoprint of high-evidence fusions vs clinical data
plots/*_circos.png	circos plot of robust fusions per cohort

resultTable.xlsx columns

column	meaning
cohort, sample	from the clinical table
caller	AR (Arriba) or FC (FusionCatcher)
gene1, gene2, label	5' and 3' partner gene and thegene1::gene2 label
reciprocal	TRUE if this is the reciprocal orientation of a call also reported the other way round
break5prime, break3prime	breakpoint coordinates
cov	breakpoint-spanning read count
mitelman_rec	number of MitelmanDB records for this fusion
PS	Promiscuity Score (mean number of distinct partners of the two genes)
tpm5prime, tpm3prime, tpmfusion	TPM of the 5' partner, 3' partner and the fusion transcript
FTS5, FTS3, FTS	Fusion Transcript Score of the 5' side, 3' side, and their mean
RS	Robustness Score (fraction of the samples calling this fusion in which FTS passes)
known	known if the fusion is a validated ChimerDB entry (>= 2 PMIDs, PCR/Sanger), else unknown
passCallFilt, passBL, passPS, passFTS, passRS	whether each criterion is met
callerOverlap	TRUE if called by both Arriba and FusionCatcher
ev_level	evidence level 0-7 (see below)
karyo, mol	agreement with theKaryotype / otherCyto column (Y / S / N)

Evidence level

ev_level counts how many of these seven criteria a call meets:

1. passCallFilt - passes the caller's built-in filters
2. passBL - not on the healthy-tissue blacklist
3. passPS - Promiscuity Score <= the highest PS among known fusions in the run (maxPS)
4. passFTS - FTS5, FTS3 >= 0.025; FTS5, FTS3 < 1; FTS >= 0.05
5. passRS - Robustness Score >= 0.5
6. callerOverlap - called by both callers
7. known - validated ChimerDB entry

ev_level >= 6 marks high-confidence candidates.

Example

Example on three public LL-100 cell lines: see example/.