Novel Statistical Methods for High-Throughput Proteomics Experiments
A team of researchers led by John Hanfelt, PhD (Emory University ACTSI/Biostatistics, Epidemiology, & Research Design [BERD]) and Junmin Peng, PhD (Department of Human Genetics, Emory University) with the assistance of Emory biostatistics doctoral student Sameera Wijayawardana have developed the software toolkit SAPHIRRA (Statistical Analysis of Protein High-Throughput Robust Relative Abundances), which provides a new set of powerful statistical methods to analyze large proteomics data sets. With SAPHIRRA, researchers can reliably estimate the relative abundances of proteins in a complex mixture consisting of possibly thousands of proteins and identify those proteins that are differentially expressed. A novel feature is that SAPHIRRA avoids many of the restrictive modeling assumptions found in many current analytical approaches; for example, SAPHIRRA does not require that the relative abundances of proteins are normally distributed. Written using the Matlab mathematical programming language, SAPHIRRA is designed to be used in experiments based on heavy/light labeling of proteins followed by LC-MS/MS (liquid chromatography -tandem mass spectrometry). The SAPHIRRA toolkit includes methods to efficiently combine information from individual peptides by smooth local averaging of reciprocal variance estimates; generation of test statistics incorporating information from both matched and unmatched proteins; visually displays of the proteins' relative abundances and test statistics; and control of the local false discovery rate. These new statistical methods have been shown to dramatically reduce the false discovery rate and yield more true discoveries than are possible with standard approaches for analyzing relative abundances of proteins in a complex mixture (Wijayawardama et al., manuscript in preparation).