Brown Statistics Seminar featuring Michael I. Love, PhD, University of North Carolina at Chapel Hill
Michael I. Love, PhD
Department of Biostatistics, Department of Genetics
University of North Carolina at Chapel Hill
“Adaptive priors for estimating effect sizes with sequence count data”
In high-throughput genomics applications, investigators often aim to determine those genomic features (genes, transcripts, peaks) with relevant changes in sequence read counts across experimental conditions, despite technical and biological variability in the observations. A common task is to accurately estimate the effect size, often in terms of a logarithmic fold change (LFC) in expression levels (RNA-seq) or binding strength (ChIP-seq). When the counts of reads are low or highly variable in either or both conditions, the maximum likelihood estimates for the LFC has high variance, leading to large estimates not representative of true differences, and poor ranking of features by effect size. This talk will discuss our introduction of an adaptive, Cauchy prior distribution for effect sizes, which avoids the use of thresholds or pseudocounts to stabilize effect sizes. The proposed method has lower bias than previously proposed shrinkage estimators, while still reducing variance for those features with little information for statistical inference.