A semi-supervised learning-based multispectral sorting method for cross-linked mass spectrometry

Abstract

The invention provides a semi-supervised learning-based multi-cross-linked-mass-spectrum sorting method. The method comprises the steps of 1) performing single spectrum matching and sorting on spectrograms to obtain corresponding optimal cross-linked dipeptide single spectrum matching results; extracting multi-spectrum matching eigenvectors of current peptide spectrum matching results, wherein dynamic features include an SVM (Support Vector Machine) score, a parent ion error ratio feature, a modification ratio feature and the like; 2) in the obtained cross-linked dipeptide matching results, selecting the results having the FDR (False Discovery Rate) in a preset FDR threshold and belonging to positive samples for constructing a positive sample library, and selecting the results of all negative samples for constructing a negative sample library; updating the multi-spectrum matching eigenvectors according to new training samples; 3) training an SVM classifier; 4) re-scoring all the cross-linked dipeptide results by using the trained SVM classifier; and 5) judging whether iteration continues to be performed or not according to a preset iteration condition, and outputting a multi-spectrum sorting result based on a current SVM score when the iteration is ended. The multi-spectrum sorting method is high in sensitivity and stable in performance.

Description

technical field [0001] The invention relates to bioinformatics technology, in particular, the invention relates to the identification technology of cross-linked mass spectrometry. Background technique [0002] At present, chemical cross-linking mass spectrometry has become the mainstream technology for studying protein structure and protein-protein interaction. In the prior art, due to the huge search space of cross-linked dipeptides, the research on computational methods is relatively lagging behind, and there is a problem of low accuracy in identifying cross-linked proteins in large-scale databases. Due to the weak normalization effect of the scoring function within a single spectrum, it is not suitable for the comparison of peptide spectrum matching between multiple spectra. Therefore, it is usually necessary to use a multispectral sorting algorithm to improve the sensitivity of cross-linked peptide identification. There are currently two types of multispectral sorting a...

AI Technical Summary

Problems solved by technology

Compared with the first method, the above-mentioned second multispectral sorting method has higher sensitivity and faster speed, but it has the disadvantage of instability, that is, the sensitivity fluctuates greatly on different data sets

Images