Time: 10:45am Tuesdays.
Level 7 Seminar Room 2, WEHI1
14 March 2017
Improved normalization of Nanostring nCounter gene expression dataRamyar Molania
The Nanostring nCounter platform is being increasingly used for research and clinical studies due to its capability to directly measure highly-multiplexed gene expression from a wide range of samples without an amplification step. Adjusting for sample loading variation and technical effects is essential for accurate interpretation of the gene expression data. We utilized a newly developed normalization method, Removing Unwanted Variation III (RUVIII) that relies on factor analysis of appropriate panels of control genes and technical replicate samples to correct for sample loading differences as well as technical variation. We showed that the current Nanostring normalization approaches that involve Nanostring positive and negative spike-in control and housekeeping genes mostly failed to remove technical variation including batch effects, particularly when the data came from large or complex experiments. Furthermore, normalization based on Nanostring spike-ins added more unwanted variation to data as the platform associated technical variation is discordant with sample related technical variation. We applied the RUVIII method to ~ 6000 samples from different in-house and published Nanostring gene expression data sets. Different statistical metrics were employed to evaluate the performance of RUVIII to determine if unwanted variation was removed. We also assessed whether biology of interest was revealed in RUVIII normalized data. The RUVIII method leads to more robust and accurate results compared to other Nanostring normalization methods, in particular when data was compromised by complex technical and batch effects. In addition, it was demonstrated that interpretation of the data was facilitated by RUVIII allowing better correspondence to the underlying biology.