DeFuse

DeFuse identifies gene fusions from RNA-Seq (whole transcriptome shotgun sequencing) data to detect somatic genomic rearrangements that alter cancer transcriptomes.

Key Features:

• Enhanced sensitivity: Considers all possible alignments and potential fusion boundary locations rather than restricting to unique best-hit alignments or known exon ends.
• Discordant paired-end and split-read integration: Integrates clusters of discordant paired-end alignments with split read alignment analysis to pinpoint fusion boundaries.
• Machine learning classifier: Incorporates an adaboost classifier trained on 11 sequence-derived features to improve specificity.
• Empirical validation and application: Classifier validated on a curated RT-PCR confirmed dataset of 60 true positive and 61 true negative fusion sequences (AUC 0.91) and applied to 40 ovarian tumor samples, one ovarian cancer cell line, and three sarcoma samples to discover novel fusions.

Scientific Applications:

• Cancer gene fusion discovery: Detection of somatic gene fusions in cancers including ovarian cancer and sarcomas using RNA-Seq data.
• Transcriptome and mutational landscape characterization: Comprehensive characterization of transcriptomes to assess how fusion events can influence gene expression and inform studies of cancer biology and potential therapeutic investigation.

Methodology:

Analyzes all possible read alignments and fusion boundary locations, clusters discordant paired-end alignments, performs split read alignment analysis to define boundaries, and applies an adaboost classifier trained on 11 features; classifier validation used 60 RT-PCR-confirmed true positives and 61 RT-PCR-confirmed true negatives (AUC 0.91).