Network alignment are computationally hard problems. - Roland Krause

Reference based indexing: A distance in 2D can be estimated by a reference point with lower and upper bound. - Roland Krause

Multiple reference points will shrink the possible space, so distance might not be required to be measured directly. - Roland Krause

Filter possible networks for comparison. - Roland Krause

Lower bound calculation doable. Upper bound calculation more difficult, fails in some specific cases. - Roland Krause

Creation of reference networks: should be small (local alignment is NP-complete), comprehensive, non-redundant. - Roland Krause

For construction: Use an existing network from the a database C, modify the network by deletion and extension, comparison to existing networks R and include if too similar. - Roland Krause

Results: KEGG gene reg network, used all > 15 nodes. Extracted 6 -8 nodes using random walks. QNet algorithm for comparison. - Roland Krause

Q: Comprehensivity of the reference set? How comprehensive can a reference network be? A: No checking after training due to reduncy removal. - Roland Krause

Q: Webserver or available implementation? A: Work as progress, query can take 5 hours. But code is available. - Roland Krause

Q: How did you change parameters to achieve different accuracies? A: Different number of references. - Roland Krause

Q: Different meanings of networks. Can this be considered in the alignment? A: Interesting question but level of abstraction does not allow currently. Could be expanded. - Roland Krause