Skip to content

Evaluation of HRR network #150

Open
proost opened this issue Feb 28, 2017 · 4 comments
Open

Evaluation of HRR network #150

proost opened this issue Feb 28, 2017 · 4 comments

Comments

@proost
Copy link
Owner

proost commented Feb 28, 2017

Script at: PhyloNet\Fig - evaluating the network

Potential implementation. Import network with HRR cutoff of 100, then run test to check optimal setting and store that value in the database filter on the fly when constructing graphs. (Would allow users to request other values if they desire to do so, however extra data that needs to be pulled from DB creates overhead)

Faster would be to iterate once over the network and remove links with hrr > cutoff.

@mutwil
Copy link
Collaborator

mutwil commented Jun 12, 2017

ATTED uses a very similar rank metric (but not identical), and they set cutoff of 100, so we could theoretically get away with 100 just by citing it.
I observed that the performance of the network peaks around HRR 300 for multiple species, but there, network density is too high! We could argue that 100 is a good compromise between performance and density.

@proost
Copy link
Owner Author

proost commented Jun 13, 2017

Did you penalize the performance-score for increasing the neighborhoods size ?

@mutwil
Copy link
Collaborator

mutwil commented Jun 13, 2017

No. After collecting the genes in the neighborhood, they are treated equally regardless of being in 1st, 2nd, 3rd...

@mutwil
Copy link
Collaborator

mutwil commented Jun 13, 2017

But as discussed, let's stick to direct (i.e. 1st) neighborhoods, as this makes our lives so much easier.

@proost proost added this to the CoNekT 2.0 milestone Jan 25, 2018
Sign in to join this conversation on GitHub.
Projects
None yet
Development

No branches or pull requests

2 participants