Evaluation and Implementation of Transcript Fusion Callers from Canine Cancer RNA-Seq Data
Pathogenic somatic fusions are abundant in many human cancers, particularly pediatric cancers. Detection of these fusions has advanced human cancer diagnosis and treatment. However, fusions are understudied in canine cancers. Canine tumors occur naturally in pet dogs, presenting clinical features similar to human cancers. Thus, new information gained from studying fusion transcripts in dog cancers is highly translatable to rare and pediatric human cancers. One barrier to studying canine cancer fusions is the lack of dog- specific analysis tools. To overcome this, human fusion calling tools were adapted and tested for use with synthetic and real RNA sequencing data from canine tumors to determine sensitivity and positive predictive value. Synthetic canine data was generated by simulating reads from a normal DNA sample. Out of the adapted fusion callers tested, one outperformed the others across most testing variables followed by a second one for synthetic data. Individually a few of the callers agreed on a majority of the true fusions while together they identified a total of 95%. One of the callers did not detect any fusions. Using real tumor data, a few of the callers had a consensus of 0.9% over all fusion calls. The fusions detected by multiple callers were manually confirmed. Variability between callers shows the need to utilize more than one algorithm to obtain confidence in the fusions called. The top performers with synthetic data were chosen to be run together as a fusion detector script to be integrated into the canine cancer analysis pipeline to advance comparative genomic studies.