Deep neural network for detecting arbitrary precision peptide features through attention based segmentation

Abstract

A promising technique of discovering disease biomarkers is to measure the relative protein abundance in multiple biofluid samples through liquid chromatography with tandem mass spectrometry (LC-MS/MS) based quantitative proteomics. The key step involves peptide feature detection in the LC-MS map, along with its charge and intensity. Existing heuristic algorithms suffer from inaccurate parameters and human errors. As a solution, we propose PointIso, the first point cloud based arbitrary-precision deep learning network to address this problem. It consists of attention based scanning step for segmenting the multi-isotopic pattern of 3D peptide features along with the charge, and a sequence classification step for grouping those isotopes into potential peptide features. PointIso achieves 98% detection of high-quality MS/MS identified peptide features in a benchmark dataset. Next, the model is adapted for handling the additional ‘ion mobility’ dimension and achieves 4% higher detection than existing algorithms on the human proteome dataset. Besides contributing to the proteomics study, our novel segmentation technique should serve the general object detection domain as well.

A streamlined platform for analyzing tera-scale DDA and DIA mass spectrometry data enables highly sensitive immunopeptidomics

“…In the case of de novo sequencing, de novo peptides can be put into a new database and an additional database search is performed to estimate the FDR of de novo peptides. Deep learning applications are employed at multiple stages throughout the platform, including MS1 isotope feature detection 24 , 25 , de novo sequencing for both DDA and DIA 23 , 26 , 27 , spectrum, retention time, and collision cross section predictions 28 – 30 (Supplementary Figs. 1 , 2 , Supplementary Table 1 ).…”

PepPre: Promote Peptide Identification Using Accurate and Comprehensive Precursors

“…The method, PepPre, is compared with pParse, RawConverter, Monocle, Decon2LS, RAPID, , MaxQuant, Dinosaur, PointIso, and three enumeration-based methods (EnumInst, EnumIW, EnumEx), as summarized in Table . Being marked as “no” in ‘coelution?’ indicates that the method only performs isotope correction to the precursors provided by the instrument and cannot report other coeluted precursor ions, which may cause significant identification loss.…”

“…The method, PepPre, is compared with pParse, 17 Raw-Converter, 16 Monocle, 32 Decon2LS, 33 RAPID, 13,14 Max-Quant, 22 Dinosaur, 24 PointIso, 26 and three enumerationbased methods (EnumInst, EnumIW, EnumEx), as summarized in Table 1. Being marked as "no" in 'coelution?'…”

“…Compared to MaxQuant, Dinosaur uses the averagine model to determine monoisotopic masses and to discard incredible features. DeepIso and PointIso apply recent artificial intelligence technologies. DeepIso builds modules based on neural networks to detect peptide features.…”

“…Instead of detecting peptide precursor ions of an MS2 scan directly, many methods − try to extract all peptide features of the LC–MS map. The methods report the mass, charge state, and retention time of each peptide feature.…”