IceR improves proteome coverage and data completeness in global and single-cell proteomics

Mathias Kalxdorf,Torsten Müller,Jeroen Krijgsveld,Oliver Stegle

doi:10.1038/s41467-021-25077-6

Abstract

Label-free proteomics by data-dependent acquisition enables the unbiased quantification of thousands of proteins, however it notoriously suffers from high rates of missing values, thus prohibiting consistent protein quantification across large sample cohorts. To solve this, we here present IceR (Ion current extraction Re-quantification), an efficient and user-friendly quantification workflow that combines high identification rates of data-dependent acquisition with low missing value rates similar to data-independent acquisition. Specifically, IceR uses ion current information for a hybrid peptide identification propagation approach with superior quantification precision, accuracy, reliability and data completeness compared to other quantitative workflows. Applied to plasma and single-cell proteomics data, IceR enhanced the number of reliably quantified proteins, improved discriminability between single-cell populations, and allowed reconstruction of a developmental trajectory. IceR will be useful to improve performance of large scale global as well as low-input proteomics applications, facilitated by its availability as an easy-to-use R-package.

Highlights

Label-free proteomics by data-dependent acquisition enables the unbiased quantification of thousands of proteins, it notoriously suffers from high rates of missing values, prohibiting consistent protein quantification across large sample cohorts
We have evaluated its performance on published plasma and single-cell proteomics data sets enabling highly increased numbers of reliably quantified proteins, increased data completeness and improved discriminability between single-cell populations allowing de novo reconstruction of a developmental trajectory
Among the most salient findings, (i) IceR could transfer peptide sequence information for nearly twice the number features due to its hybrid peptide identity propagation (PIP) approach and its superior two-step alignment algorithm (Supplementary Fig. 2b, c); (ii) false discovery rate (FDR) of direct ion current extraction (DICE)-based peak selection was estimated at 0.6% (Supplementary Fig. 2d); (iii) on average 85% of all peptide features could be directly quantified by IceR (52% by MaxQuant) (Supplementary Fig. 2a, e); (iv) IceR reduced missing value rates by 12-fold compared to MaxQuant, being at par with DeMix-Q (Supplementary Fig. 2f, g); (v) IceR resulted in most accurate and precise protein abundance ratio estimations (Supplementary Fig. 2h) and enhanced statistical power for DE analyses compared to DeMix-Q (Supplementary Fig. 2i)

Summary

Introduction

Label-free proteomics by data-dependent acquisition enables the unbiased quantification of thousands of proteins, it notoriously suffers from high rates of missing values, prohibiting consistent protein quantification across large sample cohorts. This has been successfully used to characterize thousands of proteins in countless proteomic studies, many peptides escape fragmentation due to the stochasticity of precursor selection, leading to missing data[5,6] This is caused by proteome complexity and dynamic range, and it persists despite continuous improvements of sensitivity and acquisition speeds of mass spectrometers[7]. Multiplexed labelling strategies such as TMT allow the simultaneous detection and quantification of peptides in an 11 or even 16-plexed manner at significantly reduced rates of missing values[14] This approach has been used for single-cell analysis, using one of the TMT channels to boost the signal in MS1 to benefit detection of peptides in the other channels[15,16]. Ion-based PIP is even more dependent on accurate feature alignment to enable narrow DICE windows as otherwise co-eluting species can distort quantity estimations or introduce false positives

Objectives

Methods

Results

Conclusion