Comparison and Evaluation of Clustering Algorithms for Tandem Mass Spectra

Abstract

In proteomics, liquid chromatography-tandem mass spectrometry (LC-MS/MS) is established for identifying peptides and proteins. Duplicated spectra, that is, multiple spectra of the same peptide, occur both in single MS/MS runs and in large spectral libraries. Clustering tandem mass spectra is used to find consensus spectra, with manifold applications. First, it speeds up database searches, as performed for instance by Mascot. Second, it helps to identify novel peptides across species. Third, it is used for quality control to detect wrongly annotated spectra. We compare different clustering algorithms based on the cosine distance between spectra. CAST, MS-Cluster, and PRIDE Cluster are popular algorithms to cluster tandem mass spectra. We add well-known algorithms for large data sets, hierarchical clustering, DBSCAN, and connected components of a graph, as well as the new method N-Cluster. All algorithms are evaluated on real data with varied parameter settings. Cluster results are compared with each other and with peptide annotations based on validation measures such as purity. Quality control, regarding the detection of wrongly (un)annotated spectra, is discussed for exemplary resulting clusters. N-Cluster proves to be highly competitive. All clustering results benefit from the so-called DISMS2 filter that integrates additional information, for example, on precursor mass.

Hierarchical clustering of MS/MS spectra from the firefly metabolome identifies new lucibufagin compounds

“…Here, a protocol is presented that hierarchically clusters MS/MS spectra from metabolomics data and outputs fragment ion tables in an easy to interrogate manner. Many studies have explored numerous ways in clustering MS/MS spectra for proteomics analysis and mass spectral dereplication 16,[28][29][30][31] however, clustering MS/MS data for metabolomics has considerations not generally applicable to proteomics data such as distinguishing isobaric ions that are often dereplicated in many of these analysis pipelines 4 . The BioDendro protocol does not approach dereplication in the traditional manner of combining MS/ MS data of same precursor mass and high similarity spectra regardless of retention time but rather a feature has a single alignment to an MS/MS spectrum within an m/z and retention time tolerance.…”

Large-scale tandem mass spectrum clustering using fast nearest neighbor searching

“…Next, the pairwise distance matrix is used to cluster the data using the DBSCAN algorithm (figure 1D) [8, 33]. Briefly, if a given number of spectra are close to each other and form a dense data subspace, with closeness defined relative to a user-specified distance threshold, they will be grouped in clusters.…”

Future Prospects of Spectral Clustering Approaches in Proteomics

“…Two recent studies showed considerable differences in the evaluation of spectral clustering algorithms, with regard to accuracy, and compute performance. There are several unresolved challenges in this area.…”