Statistics for Next Generation Sequencing – Meeting Report

AnAlysis of RnA-seq dAtA Profiling the transcriptome has been a central application of NGS technologies. Since the sequencing technology generates short reads, the first step is to map the reads onto the source genome, genes, and transcripts. Despite development of many algorithms and tools for mapping reads to the reference genomes, accurately mapping RNA-seq reads remains a tough problem due to the complexity of the transcriptome. Thomas Wu from Genentech presented his recent work on mapping RNA-seq sequence reads and gene structure analysis from RNA-seq data in their Genomic Short-read Nucleotide Alignment Program (GSNAP) software package. They use probabilistic models and known splicing information to guide alignment and variant/ splicing detection with consideration of all possible combinations of major and minor alleles. Wu also presented GSTRUCT, a pipeline for assembling alignment results to gene structures and predicting isoforms and gene fusion events. Unlike microarray data, sequencing data were originally thought to be digital and without need for intensive normalization. However, Zhijin Wu from Brown University demonstrated that gene expression data from RNA-seq could suffer from strong biases related to transcript length and GC content. She showed that the biases can be removed by a carefully formulated statistical model which combines a generalized linear model with a quantile normalization procedure. The clustering of RNA-seq data often relies on heuristic methods used widely in microarray data analysis. Peng Liu from Iowa State University demonstrated that a mixture of negative binomial models for clustering the RNA-seq data together with an Expectation-Maximization (EM) algorithm to obtain the model parameters can improve the performance substantially (R software package MBCluster.Seq in CRAN). In addition to the methodologically focused talks, Ali Mortazavi from UC Irvine presented preliminary results regarding RNA editing events from RNA-seq data generated by the ENCODE project. Mortazavi showed the existence of additional RNADNA sequence differences besides SNPs and canonical RNA editing events. He hypothesized that these additional sequence differences are the results of unknown technical artifacts unlike the conclusion from the work of Li et al. (2011).