|
|
Lincoln Harris is a fifth year PhD student in the Genome Sciences department at the University of Washington. Prior to graduate school, he worked on data analysis and methods development for single-cell RNA-sequencing. In Genome Science, he made the transition to proteomics and focused on developing and evaluating methods for handling missing values in quantitative proteomics. He's a member of the Noble lab but work closely with the MacCoss lab on several projects. |
Missing values in data-independent (DIA) acquisition proteomics reduce statistical power and reproducibility and make it difficult to compare across runs. Traditionally, missing values have been handled in one of two ways. The first is peptide identity propagation (PIP), in which peptide identifications from “donor” runs are transferred to unassigned MS2 features in “acceptor” runs. Popular methods for PIP include MaxQuant match-between-runs (MBR) feature and TRIC. The second way missing values are handled is “plug-in” imputation, in which statistical or machine learning methods are used to estimate missing quantitations using only the observed quantitations. Popular plug-in methods include Perseus and k-nearest neighbors (kNN).
Read More
We have developed a hybrid method that imputes retention time (RT) boundaries in a spectral library. After boundary imputation, peptide quantitation is performed with Skyline by integrating the area under the chromatographic peaks (AUC) within the imputed RT bounds. We compare RT boundary imputation to existing PIP and plug-in methods. We evaluate quantitative accuracy on three DIA datasets: (i) a matrix-matched calibration curves dataset, (ii) a large-scale Alzheimer’s disease (AD) dataset, and (iii) a dataset derived from plasma extracellular vesicles. All three datasets were accessed via Panorama. We show that RT boundary impute increases the number of differentially quantifiable peptides, reduces peptide lower limit of quantification and produces more accurate quantitations than plug-in methods.
We are currently working to integrate our RT boundary imputation method, called Nettle, into Skyline. This will provide Skyline users with an effective and principled method for reducing missingness in their DIA proteomics experiments. We stress that Nettle differs from traditional plug-in methods in that peptide quants are still derived from a measured quantity, that is, the AUC within an imputed RT window.
Read Less
