A quick intro to networks

Lightning talk from the April 25th joint meetup of R Users ATX and R Ladies ATX

Links:

Introduction to ggraph: Layouts – Thomas Lin Pederson
https://www.data-imaginist.com/2017/ggraph-introduction-layouts/

https://github.com/TheOpteProject/LGL

http://clairemcwhite.github.io/lgl-guide/

"LGL: creating a map of protein function with an algorithm for visualizing very large biological networks.
" https://www.sciencedirect.com/science/article/pii/S0022283604004851?via%3Dihub

Transcript

1. Claire McWhite @clairemcwhite 4/25/18 A quick intro to networks
2. Networks are …  First a data structure  Second a data visualization
3. A minimal network Node A Node B Edge A-B node1 node2 A B Data table =
4. Edges can have attributes node1 node2 weight relationship direction A B 5 red E A C 2 red E A D 5 blue F B C 1 blue EF C D 5 blue EF B A C D
5. Nodes can have attributes node size color A 1 red B 2 red C 3 blue D 4 blue B A C D
6. D Network layout algorithms node1 node2 A B A C A D B C C D B A C Network data structure text file Visualized network Network layout algorithm Calculates node coordinates Not very interpretable by humans Ideally more interpretable by humans
7. Introduction to ggraph: Layouts – Thomas Lin Pederson https://www.data-imaginist.com/2017/ggraph-introduction-layouts/
8. Hierarchical layouts Introduction to ggraph: Layouts – Thomas Lin Pederson https://www.data-imaginist.com/2017/ggraph-introduction-layouts/ Yfiles circular layouts
9. I use mostly force-directed layouts
10. node1 node2 weight A B 1 B C 10 A B C A B C A B C 1 10 Edges act as springs in a force-directed layout
11. CC-BY martinandjean.ch 2016
12. CC-BY martinandjean.ch 2016
13. There is a sweet spot of network size for most network layout algorithms. Then you enter the hairball zone https://www.systemsbiology.org/news/2012/12/19/combing-the-hairball/
14. When networks get too large, they may become uninterpretable hairballs
15. Steps to break up hairball networks Apply a threshold for edge weights. ------------------------ Thresholding -------------------
16. Steps to break up hairball networks Apply a clustering algorithm to network data ------------------------ Clustering -------------------
17. My data: No clustering 10 million edge, 1 million node network, Large Graph Layout (LGL)
18. MCL clustering strength 3
19. MCL clustering strength 4
20. MCL clustering strength 5
21. MCL clustering strength 8
22. Software  R – ggraph/tidygraph  Python – NetworkX  Gui -Cytoscape (lots of tools for biology)  Gui – Gephi  Gui - Graphviz
23. Large graph layout - LGL Protein homology graph – Edward Marcotte and Alex Adai - MOMA • My fav • Minimizes hairballness of larger networks • Scalable (handles over 1 million node networks) Adai 2004, LGL: creating a map of protein function with an algorithm for visualizing very large biological networks.
24. Post-talk addendum  It would be great if someone could make a LGL network layout algorithm plugin for cytoscape or any network visualization software.  The program currently only runs on Linux  https://github.com/TheOpteProject/LGL  My user guide -> http://clairemcwhite.github.io/lgl-guide/  Original paper describing algorithm: https://www.sciencedirect.com/science/article/pii /S0022283604004851?via%3Dihub
25. If your data looks like this at all, throw it in a network layout. node1 node2 A B A C A D B C C D (but don’t overinterpret any one layout, and remember that network layouts are fickle and minor changes to thresholds and data input choices can change everything)