Graph visualization with latent variable models

Parkkinen, Juuso; Nybo, Kristian; Peltonen, Jaakko; Kaski, Samuel

doi:10.1145/1830252.1830265

“…Since specXplore contains a large network analysis component, the choice of feature positioning in its general overview is in part a question of network layout choice. Laying out dense networks is a difficult task often addressed using force-directed layout algorithms owing to their computational tractabiity. − However, the latter do not scale well to large and dense networks and tend to produce hard-to-read or unintelligable networks rendition. , This is particularly because of often created dense “hairballs” of nodes and edges, as well as edge-crossings. − In specXplore, this is addressed by a combination of latent variable space embedding of nodes with interactively triggered partial network drawings. ,,, Here, the latent variable embedding serves as a jumping board for localized network explorations. This approach is in line with Shneiderman’s mantra of “Overview first; details on demand” .…”

Section: Discussionmentioning

confidence: 99%

“… 57 − 59 In specXplore, this is addressed by a combination of latent variable space embedding of nodes with interactively triggered partial network drawings. 37 , 39 , 60 , 61 Here, the latent variable embedding serves as a jumping board for localized network explorations. This approach is in line with Shneiderman’s mantra of “Overview first; details on demand”.…”

Section: Discussionmentioning

confidence: 99%

Tailored Mass Spectral Data Exploration Using the SpecXplore Interactive Dashboard

Mildau,

Ehlers,

Oesterle

et al. 2024

Anal. Chem.

4

0

View full text Add to dashboard Cite

Untargeted metabolomics promises comprehensive characterization of small molecules in biological samples. However, the field is hampered by low annotation rates and abstract spectral data. Despite recent advances in computational metabolomics, manual annotations and manual confirmation of in-silico annotations remain important in the field. Here, exploratory data analysis methods for mass spectral data provide overviews, prioritization, and structural hypothesis starting points to researchers facing large quantities of spectral data. In this research, we propose a fluid means of dealing with mass spectral data using specXplore, an interactive Python dashboard providing interactive and complementary visualizations facilitating mass spectral similarity matrix exploration. Specifically, specXplore provides a two-dimensional t-distributed stochastic neighbor embedding embedding as a jumping board for local connectivity exploration using complementary interactive visualizations in the form of partial network drawings, similarity heatmaps, and fragmentation overview maps. SpecXplore makes use of state-of-the-art ms2deepscore pairwise spectral similarities as a quantitative backbone while allowing fast changes of threshold and connectivity limitation settings, providing flexibility in adjusting settings to suit the localized node environment being explored. We believe that specXplore can become an integral part of mass spectral data exploration efforts and assist users in the generation of structural hypotheses for compounds of interest.

show abstract

“…In the worst case, naive attempts to visualize such networks as (force-directed) straight-line node-link diagrams result in so-called “hairballs”; dense, unintelligible clusters of nodes and edges, which do not allow for meaningful exploration of either local or global topology [60, 61]. In specXplore, we tackle the layout problem by combining a latent variable space embedding [35, 62] of nodes with interactively triggered partial network drawings [45, 63]. The latent variable embedding serves the purpose of a fixed jumping board for localized explorations.…”

Section: Discussionmentioning

confidence: 99%

Tailored mass spectral data exploration using the specXplore interactive dashboard

Mildau,

Ehlers,

Oesterle

et al. 2023

Preprint

0

View full text Add to dashboard Cite

Untargeted metabolomics promises comprehensive characterization of small molecules in biological samples. However, the field is hampered by low annotation rates and abstract spectral data. Despite recent advances in computational metabolomics, manual annotations and manual confirmation of in-silico annotations remain important in the field. Here, exploratory data analysis methods for mass spectral data provide overviews, prioritization, and structural hypothesis starting points to researchers facing large quantities of spectral data. In this research, we propose a fluid means of dealing with mass spectral data using specXplore, an interactive python dashboard providing interactive and complementary visualizations facilitating mass spectral similarity matrix exploration. Specifically, specXplore provides a two dimensional t-SNE embedding as a jumping board for local connectivity exploration using complementary interactive visualizations in the form of partial network drawings, similarity heatmaps, and fragmentation overview maps. SpecXplore makes use of of state of the art ms2deepscore pairwise spectral similarities as a quantitative backbone, while allowing fast changes of threshold and connectivity limitation settings, providing flexibility in adjusting settings to suit the localized node environment being explored. We believe that specXplore can become an integral part in mass spectral data exploration efforts and assist users in the generation of structural hypotheses for compounds of interest.Technical TermsA network is a collection of connected features. In our case, a network consists of MS/MS spectral features connected provided their spectral similarity is high. Networks are represented using node-link-diagrams.Node-link diagram -a term commonly used to refer to the graphical representation of a network via nodes and links (i.e. edges). In this paper, we use node-link diagram and network-view interchangeably.A node is a feature in a network that can be connected to other features via edges. An alternative term for node is vertex.An edge is a connection between two nodes. Other terms for edges are links or vertices.Network layout refers to the spatial arrangement of nodes and edges on an usually two dimensional plotting surface. Network layout is also sometimes referred to as embedding. This term is avoided in this paper to avoid confusion with embedding in the machine learning sense.Given a networkG(V, E), whereVdenotes its nodes andEits (weighted) edges, we define its topology as the relationships between individual (groups of) nodes and edges or the network as a whole, irrespective of the network’s layout.Molecular Networking (MN) is an exploratory data analysis technique merging spectral similarity-based topological clustering and visualization as node-link diagrams.The plain English words group/grouping are wherever appropriate to avoid jargon terms such as clustering (as in k-medoid or k-means clustering), embedding (as in projection of groups of features into a close-by lower dimensional space), or molecular families. The latter are groups of spectral data features clustered and visualized as network-views via traditional MN or feature based molecular networking (FBMN). Molecular families, usually represent smaller, disconnected networks that are part of a larger dataset. When we refer to this disconnected nature, we use the phrasing disjoint sub-network for emphasis.

show abstract

“…5). The last three graphs are laid out as an unconstrained 2D graph by a recent nodeneighborhood preserving layout method [18]. For all graphs, edge bundles were created by a d3.js plugin implementing the algorithm [20] adapted to splines.…”

Section: Methodsmentioning

confidence: 99%

Peacock Bundles: Bundle Coloring for Graphs with Globality-Locality Trade-Off

Peltonen

¹

,

Lin

²

2016

Lecture Notes in Computer Science

Self Cite

1

0

View full text Add to dashboard Cite

Bundling of graph edges (node-to-node connections) is a common technique to enhance visibility of overall trends in the edge structure of a large graph layout, and a large variety of bundling algorithms have been proposed. However, with strong bundling, it becomes hard to identify origins and destinations of individual edges. We propose a solution: we optimize edge coloring to differentiate bundled edges. We quantify strength of bundling in a flexible pairwise fashion between edges, and among bundled edges, we quantify how dissimilar their colors should be by dissimilarity of their origins and destinations. We solve the resulting nonlinear optimization, which is also interpretable as a novel dimensionality reduction task. In large graphs the necessary compromise is whether to differentiate colors sharply between locally occurring strongly bundled edges ("local bundles"), or also between the weakly bundled edges occurring globally over the graph ("global bundles"); we allow a user-set global-local tradeoff. We call the technique "peacock bundles". Experiments show the coloring clearly enhances comprehensibility of graph layouts with edge bundling.

show abstract

Graph visualization with latent variable models

Cited by 5 publications

References 28 publications

Tailored Mass Spectral Data Exploration Using the SpecXplore Interactive Dashboard

Tailored Mass Spectral Data Exploration Using the SpecXplore Interactive Dashboard

Tailored mass spectral data exploration using the specXplore interactive dashboard

Peacock Bundles: Bundle Coloring for Graphs with Globality-Locality Trade-Off

Contact Info

Product

Resources

About