Title: Noise model and statistical significance in MPSS transcription profiling
Speaker: Gustavo Stolovitzky, IBM T.J. Watson Research Center
Date: Wednesday, April 14, 2004, 1:00 pm
Location: Hill Center, Room 260, Rutgers University, Busch Campus, Piscataway, NJ
The last decade witnessed a shift in methodologies, in which high throughput technologies to probe the cell's transcriptome changed the landscape of biological research. The power of high throughput methodologies, however, contains the seed for some of its own drawbacks. Along with the many measurements that one experiment yields, there exist many spurious genes whose expression level is the by-product of high levels of noise in the technology. Thus, the quality of the analysis resulting from these high throughput assays depends crucially on our ability to understand how noise affects measurements. One of the most recently developed high throughput transcription technology is the Massively Parallel Signature Sequencing (MPSS). This technology is complementary to the gene expression array in that it does not require to know a priori what genes are expected to be transcribed. There are a number of non-trivial steps that have to be followed in MPSS from the extraction of the total RNA from a set of cells to the quantitation of transcripts. Even though a physical description of these steps is in principle possible, there usually are sequence and sample dependent noise sources that are rather difficult to model a priori. In this talk I will present an empirical noise description valid for MPSS that can be used as the null hypothesis in a hypothesis test. We will apply this test in the context of pairwise measurements and in the context of time traces.
Seminar sponsored by DIMACS/BIOMAPS Seminar Series on Quantitative Biology and Epidemiology.