 |
|
| FriendFeed is the easiest way to share online. Learn more » |
ISMB/ECCB 2011:
ISMB/ECCB
PT17: Regev Schweiger - Generative Probabilistic Models for Protein-Protein Interaction Networks – The Biclique Perspective
July 17, 2011
- Comment
- Like
- Share
Ruchira S. Datta and Roland Krause liked this
Introduction by Rob Russell: The subject Networks were second in terms of submitted papers.
- Roland Krause
Protein don't work alone. TAP, Y2H interactomes in yeast. Datebases.
- Roland Krause
(Hairball)
- Roland Krause
Introduction to random graphs (Erdös-Renyi, geometric)
- Roland Krause
Degree distribution and clustering are not described by the above models. Alternatives are preferential attachment (PA) and duplication-divergence (DD).
- Roland Krause
How to test a graph model against a real world network. Compare distribution, network motifs, dense subgraphs are used.
- Roland Krause
Contribution: maximal bicliques. Bicliques are naturally related to the DD model
- Roland Krause
Introduced for web graphs, found in yeast interactome.
- Roland Krause
DIP set, plot all bicliques in heat map. Large range.
- Roland Krause
Generate many models, count maximal biclique count, compare to the real network, compare by sum of difference of logs.
- Roland Krause
Should become independent of parameters: scan the range of parameters and select the best for the real data.
- Roland Krause
DD is similar, larger cliques, outperforms PA. Seeds are important.
- Roland Krause
More simulations, best fit with inverse geometric random seed model. DD still outperforms PA.
- Roland Krause
Most of the bicliques are in the seed of the PA. In DD, its only 5.1%. Removing the seed bicliques shows stronger effects.
- Roland Krause
Q: Influence of abundance? A: Should be independent.
- Roland Krause
Q: Have you tested other measures? A: Yes (not presented). DD gives best results.
- Roland Krause
Q: Did you use only Y2H data? A: Yes, should be investigated more.
- Roland Krause
Q: Single large scale experiments rather then DIP? A: No.
- Roland Krause
Q&A (Rob Russell) Biological relevance? The DD network model makes more sense. [...]
- Roland Krause
Q: Inclusion of smaller cliques, overcounting. Term is not right as nodes are connected with one clique? A: Hmm, yes chould be named differently.
- Roland Krause
Q: Sequence similarity? A: Others have used this to come up with binding motives.
- Roland Krause
Q: Known quality of PPI - disassociativiy of hubs, other interaction network such as TFs? A: No, tested it a bit, few organisms have larger networks, TF networks are directed.
- Roland Krause
Q; Distribution of the counts, is the average a good measure? A: It's different for each model and set of parameters, variance rather small.
- Roland Krause
There's a poster.
- Roland Krause