Rescuing discarded spectra: Full comprehensive analysis of a minimal proteome

Abstract

A common problem encountered when performing large-scale MS proteome analysis is the loss of information due to the high percentage of unassigned spectra. To determine the causes behind this loss we have analyzed the proteome of one of the smallest living bacteria that can be grown axenically, Mycoplasma pneumoniae (729 ORFs). The proteome of M. pneumoniae cells, grown in defined media, was analyzed by MS. An initial search with both Mascot and a species-specific NCBInr database with common contaminants (NCBImpn), resulted in around 79% of the acquired spectra not having an assignment. The percentage of non-assigned spectra was reduced to 27% after re-analysis of the data with the PEAKS software, thereby increasing the proteome coverage of M. pneumoniae from the initial 60% to over 76%. Nonetheless, 33,413 spectra with assigned amino acid sequences could not be mapped to any NCBInr database protein sequence. Approximately, 1% of these unassigned peptides corresponded to PTMs and 4% to M. pneumoniae protein variants (deamidation and translation inaccuracies). The most abundant peptide sequence variants (Phe-Tyr and Ala-Ser) could be explained by alterations in the editing capacity of the corresponding tRNA synthases. About another 1% of the peptides not associated to any protein had repetitions of the same aromatic/hydrophobic amino acid at the N-terminus, or had Arg/Lys at the C-terminus. Thus, in a model system, we have maximized the number of assigned spectra to 73% (51,453 out of the 70,040 initial acquired spectra). All MS data have been deposited in the ProteomeXchange with identifier PXD002779 (http://proteomecentral.proteomexchange.org/dataset/PXD002779).

Network‐based characterization of the synaptic proteome reveals that removal of epigenetic regulator Prmt8 restricts proteins associated with synaptic maturation

“…Even for the same sample, re-running it multiple times will lead to acquisition of spectra originating from different peptides/proteins. Mapping raw spectra against theoretical protein sequence libraries is also problematic: in practice, only a small part of acquired spectra are confidently assigned to peptides (Lluch-Senar et al 2016). Moreover, running different library-search algorithms may result in different proteins being identified, especially among low confidence proteins (low unique peptide support) (Nesvizhskii 2007).…”

Rapid protein-protein interaction network creation from multiple sequence alignments with Deep Learning

“…For a set of 1000 proteins (the order of the smallest known proteomes [18]), this results in 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑎𝑖𝑟𝑠 = 1000 • 999/2 = 999'000/2 = 499'500 𝑝𝑎𝑖𝑟𝑠 The difference between the full and reduced procedure for 1000 proteins is therefore:…”