Bioinformatics Seminar
Time:
Venue: Na
3 September 2019
NaXenoSplit: distinguishing human from mouse in RNA-seq from mouse xenograft models
Gordon SmythWEHI Bioinformatics
In human tumour xenografts ; human tumour cells are transplanted into immunocompromised mice that do not reject human cells. Depending upon the number of cells injected ; or the size of the tumour transplanted ; the tumour will develop in the mouse over 1-8 weeks and the response to appropriate therapeutic regimes can be studied in vivo. A particular type of xenograft that is significant for personalized medicine is the patient-derived xenograft (PDX) ; whereby fresh tumour cells from a human patient are transplanted into the immunocompromised mice. PDXs allow us to profile tumour cells ; to compare metasticized tumours to the original tumour ; or to profile circulating tumour cells (CTCs). The classic bioinformatics problem associated with mouse xenograft models is that samples taken from xenograft models may contain a mixture of mouse and human cells ; but only the human cells are of interest. So we need to separate the human sequence reads from the mouse reads when processing RNA-seq or DNA-seq libraries from xenografts. In this talk ; we develop some efficient methodology and an R package to separate human RNA sequence reads from mouse reads. We start with an Rsubread facility that writes mouse and human BAM files in parallel. Then we use logistic regression to distinguish human from mouse ; using the number of perfect-match aligned bases ; mismatch aligned bases ; insertions and deletions as covariates. We explore the influence of read length on the optimal logistic regression predictor. We apply this software to PDX samples on metasticized cancers and CTCs. We explore the limiting case of CTCs ; when the proportion of human RNA in the samples is typically very small. This is joint work with Dr Goknur Giner and uses data from the Visvader/Lindeman Lab.;;;;