Scalable graph analysis tools for the connectomics community

Matelsky, Jordan; Johnson, Erik C.; Wester, Brock A.; Gray-Roncal, William

doi:10.1101/2022.06.01.494307

Cited by 2 publications

(4 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The ellipsoid body graph took a total of 1,125 CPU-days to run on the compute cluster (around 36,000 searches at µ = 45 minutes each). This runtime is dominated by the edge count and local topology of the host graph, rather than its vertex count, and so to our knowledge there is no straightforward way to estimate runtime prior to execution at this time [15]. To further reduce the wallclock execution time, we also provide a cloud-based software implementation that uses serverless compute (i.e., Amazon Web Services Lambda and DynamoDB) [15].…”

Section: Methodsmentioning

confidence: 99%

“…This runtime is dominated by the edge count and local topology of the host graph, rather than its vertex count, and so to our knowledge there is no straightforward way to estimate runtime prior to execution at this time [15]. To further reduce the wallclock execution time, we also provide a cloud-based software implementation that uses serverless compute (i.e., Amazon Web Services Lambda and DynamoDB) [15]. All results from both of these connectome graph searches, as well as our randomized network results, will be made publicly available.…”

Section: Methodsmentioning

confidence: 99%

“…Furthermore, connectomes are often one-of-a-kind, meaning that traditional statistical models are challenging or impossible to implement. Finally, exhaustive subgraph matching or counting can be computationally intractable for even relatively small connectomes [15, 16].…”

Section: Introductionmentioning

confidence: 99%

“…And, of course, there are a number of additional heuristics that can be used to improve computational feasibility. Some of the popular approaches include MAVisto, MFinder, FANMOD, and MODA [21, 22, 23, 24, 25], as well as our more recent contributions, DotMotif and GrandIso [10] and their cloud-scale equivalents [15].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Data-driven motif discovery in biological neural networks

Matelsky,

Robinette,

Wester

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Data from a variety of domains are represented as graphs, including social networks, transportation networks, computer networks, and biological networks. A key question spans these domains: are there meaningful repeated subgraphs, or motifs, within the structure of these larger networks? This is a particularly relevant problem when searching for repeated neural circuits in networks of biological neurons, as the field now regularly produces large brain connectivity maps of neurons and synapses, or connectomes. Given acquisition costs, however, these neuron-synapse connectivity maps are mostly one-of-a-kind. With current graph analysis techniques, it is very challenging to discover new "interesting" subgraphs a priori given small sample sizes of host graphs. Another challenge is that for even relatively modest graph sizes, an exhaustive search of all possible subgraphs is computationally intractable. For these reasons, motif discovery in biological graphs remains an unsolved challenge in the field. In this work, we present a motif discovery approach that can derive a list of undirected or directed motifs, with occurrence counts which are statistically significant compared to randomized graphs, from a single graph example. We first address common pitfalls in the current most common approaches when testing for motif statistical significance, and outline a strategy to ameliorate this problem with improved graph randomization techniques. We then propose a progressive-refinement approach for motif discovery, which addresses issues of computational cost. We demonstrate that our sampling correction technique allows for significance testing of target motifs while highlighting misleading conclusions from standard random graph models. Finally, we share our reference implementation, which is available as an open-source Python package, and demonstrate real-world preliminary results on the C. elegans connectome and the ellipsoid body of the Drosophila melanogaster fruit fly connectome.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Data-driven motif discovery in biological neural networks

Matelsky,

Robinette,

Wester

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

Beyond neurons: computer vision methods for analysis of morphologically complex astrocytes

Syed,

Youssef,

Schober

et al. 2024

Front. Comput. Sci.

View full text Add to dashboard Cite

The study of the geometric organization of biological tissues has a rich history in the literature. However, the geometry and architecture of individual cells within tissues has traditionally relied upon manual or indirect measures of shape. Such rudimentary measures are largely a result of challenges associated with acquiring high resolution images of cells and cellular components, as well as a lack of computational approaches to analyze large volumes of high-resolution data. This is especially true with brain tissue, which is composed of a complex array of cells. Here we review computational tools that have been applied to unravel the cellular nanoarchitecture of astrocytes, a type of brain cell that is increasingly being shown to be essential for brain function. Astrocytes are among the most structurally complex and functionally diverse cells in the mammalian body and are essential partner cells of neurons. Light microscopy does not allow adequate resolution of astrocyte morphology, however, large-scale serial electron microscopy data, which provides nanometer resolution 3D models, is enabling the visualization of the fine, convoluted structure of astrocytes. Application of computer vision methods to the resulting nanoscale 3D models is helping reveal the geometry and organizing principles of astrocytes, but a complete understanding of astrocyte structure and its functional implications will require further adaptation of existing computational tools, as well as development of new approaches.

show abstract

Scalable graph analysis tools for the connectomics community

Cited by 2 publications

References 18 publications

Data-driven motif discovery in biological neural networks

Data-driven motif discovery in biological neural networks

Beyond neurons: computer vision methods for analysis of morphologically complex astrocytes

Contact Info

Product

Resources

About