Poster Presentation 25th Annual Lorne Proteomics Symposium 2020

A statistical approach to rank differentially expressed proteins using confidence bound on effect size. (#141)

Anup D Shah 1 , Paul F Harrison 2 , Ralf B Schittenhelm 1 , David R Powell 2
  1. Monash Proteomics and Metabolomics Facility, Monash University, Clayton, VIC, Australia
  2. Monash Bioinformatics Platform, Monash University, Clayton, VIC, Australia

High-throughput quantitative “omics” experiments often involve characterisation of a large number of biomolecules. Over the years, prioritising this list for subsequent follow-up analysis is generally carried out by applying arbitrary p value and/or log fold change thresholds. Recently, statisticians worldwide have expressed their concerns over the use of p value as a surrogate for effect size (1). Additionally, filtering by fold changes also requires an expert insight to determine the threshold. As an alternative, a method of ranking biomolecules by confidence bounds on the effect size, ‘topconfects’, has been proposed recently (2). It emphasises on biggest effect size in contrast to consistent but small changes in case of p value based filtering. This method has shown promising results with ‘transcriptomics’ data highlighting the biologically significance. Here, we applied ‘topconfects’ to a label-free proteomics dataset and found marked differences in top-ranked protein list generated by confidence bounds and p values.

  1. Ronald L. Wasserstein & Nicole A. Lazar (2016) The ASA Statement on p-Values: Context, Process, and Purpose, The American Statistician, 70:2, 129-133
  2. Harrison, P.F., Pattison, A.D., Powell, D.R. Beilharz T.H. (2019) Topconfects: a package for confident effect sizes in differential expression analysis provides a more biologically useful ranked gene list. Genome Biol 20, 67