Robust unsupervised deconvolution of linear motifs characterizes 68 protein modifications at proteome scale

Theodore G Smith,Philipp F Lange,Anuli C Uzozie,Siyuan Chen

doi:10.1038/s41598-021-01971-3

Theodore G Smith, Philipp F Lange + Show 2 more

Open Access

https://doi.org/10.1038/s41598-021-01971-3

Copy DOI

Abstract

The local sequence context is the most fundamental feature determining the post-translational modification (PTM) of proteins. Recent technological improvements allow for the detection of new and less prevalent modifications. We found that established state-of-the-art algorithms for the detection of PTM motifs in complex datasets failed to keep up with this technological development and are no longer robust. To overcome this limitation, we developed RoLiM, a new linear motif deconvolution algorithm and webserver, that enables robust and unbiased identification of local amino acid sequence determinants in complex biological systems demonstrated here by the analysis of 68 modifications found across 30 tissues in the human draft proteome map. Furthermore, RoLiM analysis of a large-scale phosphorylation dataset comprising 30 kinase inhibitors of 10 protein kinases in the EGF signalling pathway identified prospective substrate motifs for PI3K and EGFR.

Highlights

The local sequence context is the most fundamental feature determining the post-translational modification (PTM) of proteins
The state of the proteome is shaped by post-translational modification affecting the majority of proteins in one form or another[1,2]
Mass spectrometry allows for the de-novo detection of unknown modifications and technological advancements in enrichment procedures, top-down approaches and instrument sensitivity, have led to a steep increase in the modifications known to occur in vivo[3,4]

Summary

Introduction

The local sequence context is the most fundamental feature determining the post-translational modification (PTM) of proteins. When we analyzed the local sequence context of protein modifications identified in the TagGraph dataset we noticed that residues found as enriched by rmotif-X were not distributed around the modified central residue. When we analyzed the same dataset with randomly reordered input sequences using MoMo default settings, only 30% of identified patterns were identical between original and reordered sequence order (Fig. 1b, Supplementary Data S2).

Results

Conclusion