Density-Based Clustering Based on Hierarchical Density Estimates

Campello, Ricardo J. G. B.; Moulavi, Davoud; Sander, Jöerg

doi:10.1007/978-3-642-37456-2_14

Cited by 1,497 publications

(1,018 citation statements)

References 19 publications

Supporting

Mentioning

887

Contrasting

Unclassified

Order By: Relevance

“…The NbClust package requires as a parameter the maximum number of clusters to look for, which is set max.nc = 30. For the extraction of clusters from the HDBSCAN hierarchy, we adopt the EOM-optimization [5] and for the nonparametric Bayesian method DP-means, we extended the R-script which is available on GitHub 6 . Table 2, which either estimate the number of clusters directly, or provide a clustering vector without any knowledge of the number of clusters.…”

Section: Discussionmentioning

confidence: 99%

“…The graph-analogue of DBSCAN has been presented in [5] and dynamically adjusting the density level , the nested hierarchical sequence of clusterings results to the HDBSCAN algorithm [5]. OPTICS [2] allows for determining the number of clusters in a dataset by counting the "dents" of the OPTICS reachability plot.…”

Section: Related Workmentioning

confidence: 99%

“…The OPTICS-ξ algorithm [2] requires an extra parameter ξ, which has to be manually set in order to find "dents" in the OPTICS reachability plot. The automatic extraction of clusters from the OPTICS reachability plot, as an extension of the OPTICS-ξ algorithm, has been presented in [27] and has been outperformed by HDBSCAN-EOM [5] in several datasets. We will examine whether some of these density based algorithms perform well on the news clustering problem and we shall compare them with our DBSCAN-Martingale, which is a modification of DBSCAN, where the density level is a random variable and the clusters are progressively extracted.…”

Section: Related Workmentioning

confidence: 99%

“…We would tackle the news categorization problem as a supervised classification problem, because training sets are available, contrary to the news clustering problem where, for example, the topic "Malaysia Airlines Flight 370" had no training set before the 8 th of March 2014. We assume that 2000 news articles is a reasonable upper bound for the number of recent news articles that can be considered for news clustering, in line with other datasets that were used to evaluate similar methods [25,5]. In all datasets (Table 2) we extract uni-grams and bi-grams, assuming a Bag-of-Words representation of text.…”

Section: Dataset Descriptionmentioning

confidence: 99%

See 3 more Smart Citations

A Hybrid Framework for News Clustering Based on the DBSCAN-Martingale and LDA

Gialampoukidis

Vrochidis

Kompatsiaris

2016

Machine Learning and Data Mining in Pattern Recognition

View full text Add to dashboard Cite

Abstract. Nowadays there is an important need by journalists and media monitoring companies to cluster news in large amounts of web articles, in order to ensure fast access to their topics or events of interest. Our aim in this work is to identify groups of news articles that share a common topic or event, without a priori knowledge of the number of clusters. The estimation of the correct number of topics is a challenging issue, due to the existence of "noise", i.e. news articles which are irrelevant to all other topics. In this context, we introduce a novel density-based news clustering framework, in which the assignment of news articles to topics is done by the well-established Latent Dirichlet Allocation, but the estimation of the number of clusters is performed by our novel method, called "DBSCAN-Martingale", which allows for extracting noise from the dataset and progressively extracts clusters from an OPTICS reachability plot. We evaluate our framework and the DBSCAN-Martingale on the 20newsgroups-mini dataset and on 220 web news articles, which are references to specific Wikipedia pages. Among twenty methods for news clustering, without knowing the number of clusters k, the framework of DBSCAN-Martingale provides the correct number of clusters and the highest Normalized Mutual Information.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Dataset Descriptionmentioning

confidence: 99%

See 2 more Smart Citations

A Hybrid Framework for News Clustering Based on the DBSCAN-Martingale and LDA

Gialampoukidis

Vrochidis

Kompatsiaris

2016

Machine Learning and Data Mining in Pattern Recognition

View full text Add to dashboard Cite

show abstract

“…In order to test if other clustering algorithms can be used to group peaks within single peak list into spin system clusters, a recently developed variation of DBSCAN called hierarchical DBSCAN (HDBSCAN) was used [110], [111]. This clustering algorithm was chosen, because it has several advantages over other clustering algorithms:…”

Section: Comparison To Hierarchical Dbscan Algorithmmentioning

confidence: 99%

Algorithms for automated assignment of solution-state and solid-state protein NMR spectra.

Smelter¹

View full text Add to dashboard Cite

Protein nuclear magnetic resonance spectroscopy (Protein NMR) is an invaluable analytical technique for studying protein structure, function, and dynamics. There are two major types of NMR spectroscopy that are used for investigation of protein structuresolution-state and solid-state NMR. Solution-based NMR spectroscopy is typically applied to proteins of small and medium size that are soluble in water. Solid-state NMR spectroscopy is amenable for proteins that are insoluble in water.In the vast majority NMR-based protein studies, the first step after experiment optimization is the assignment of protein resonances via the association of chemical shift values to specific atoms in a protein macromolecule. Depending on the quality of the spectra, a manual protein resonance assignment process often requires a considerable amount of time, from weeks to months-worth of effort even, by an experienced NMR spectroscopist.vii The resonance assignment processes for solution-state and solid-state protein NMR studies are conceptually similar, but have distinct differences due to the utilization of different NMR experiments and to the use of different resonances for grouping peaks into spin systems.Currently, there is a shortage of robust, effective software tools that can perform solid-state protein resonance assignment and there is no general software that can perform both solution-state and solid-state protein resonance assignment in a reliable, automated fashion. Hence, the motivation of this research is to design and implement algorithms and software tools that will automate the resonance assignment problem.As a result of this research, several algorithms and software packages that aid several important steps in the protein resonance assignment process were developed. For example, the nmrstarlib software package can access and utilize data deposited in the NMR-STAR format; the core of this library is the lexical analyzer for NMR-STAR syntax that acts as a generator-based state-machine for token processing. The jpredapi software package provides an easy-to-use API to submit and retrieve results from secondary structure prediction server. The single peak list and pairwise peak list registration algorithms address the problem of multiple sources of variance within single peak list and between different peak lists and is capable of calculating the match tolerance values necessary for spin system grouping. The single peak list and pairwise peak list grouping algorithms are based on the well-known DBSCAN clustering algorithm and are designed to group peaks into spin systems within single peak list as well as between different peak lists.viii

show abstract

Smyd2 conformational changes in response to p53 binding: role of the C‐terminal domain

2019

View full text Add to dashboard Cite

Smyd2 lysine methyltransferase regulates monomethylation of histone and nonhistone lysine residues using S‐adenosylmethionine cofactor as the methyl donor. The nonhistone interactors include several tumorigenic targets, including p53. Understanding this interaction would allow the structural principles that underpin Smyd2‐mediated p53 methylation to be elucidated. Here, we performed μ‐second molecular dynamics (MD) simulations on binary Smyd2‐cofactor and ternary Smyd2‐cofactor‐p53 peptide complexes. We considered both unmethylated and monomethylated p53 peptides (at Lys370 and Lys372). The results indicate that (a) the degree of conformational freedom of the C‐terminal domain of Smyd2 is restricted by the presence of the p53 peptide substrate, (b) the Smyd2 C‐terminal domain shows distinct dynamic properties when interacting with unmethylated and methylated p53 peptides, and (c) Lys372 methylation confines the p53 peptide conformation, with detectable influence on Lys370 accessibility to the cofactor. These MD results are therefore of relevance for studying the biology of p53 in cancer progression.

show abstract

Density-Based Clustering Based on Hierarchical Density Estimates

Cited by 1,497 publications

References 19 publications

A Hybrid Framework for News Clustering Based on the DBSCAN-Martingale and LDA

A Hybrid Framework for News Clustering Based on the DBSCAN-Martingale and LDA

Algorithms for automated assignment of solution-state and solid-state protein NMR spectra.

Smyd2 conformational changes in response to p53 binding: role of the C‐terminal domain

Contact Info

Product

Resources

About