Slow cna() -- too slow?

[rydog]

This is a followup to issue #315. I'm using the same system: * R version 3.2.1 (2015-06-18), x86_64-apple-darwin10.8.0 (64-bit) * OSX 10.8.5 * bio3d package: Version: 2.2-3 Date: 2015-09-04

My test case is a 451 residue protein, so my correlation matrix that I'm inputting to cna is 451x451. I'm running through the "Protein Structure Networks" tutorial: library(bio3d) library(igraph) pdb <- read.pdb( 'test.pdb') modes <- nma(pdb) cij <- dccm(modes, ncore=1) net <- cna(cij, cutoff.cij=0.3, ncore=1)

I can plot cij and it looks meaningful.

I assume I need to run cna with ncore=1 (I'm also testing it without ncore=1), but I don't know how long is reasonable to wait for the program to complete. It's been about eighty minutes, and from watching the process (in terms of CPU/memory usage) it looks like something meaningful is happening but there's no output.

I could complete this with an even-smaller fragment (around forty residues) and the HIV protein example looks like it's around 200 residues, so any suggestions how I should proceed? Are the flags that will make the cna output more verbose/incremental?

[Xinqiu Yao]

Hi,

cna() usually takes seconds to complete and eighty minutes waiting time is definitely odd. One possibility is that the network has too many edges even after using the cutoff 0.3. Can you check the cij with the function pymol.dccm()? (e.g. pymol.dccm(cij, pdb=pdb, step=0.1, type='launch')). Toggle different cij cutoff options and choose the one that looks the most reasonable (e.g. not so many edges connecting distant residues). Then, use the chosen cutoff to build network again, cna(cij, cutoff.cij=your_new_cutoff).

[rydog]

Thanks. I ran the HIV example as a test, and it does complete quickly. It seems like I just need to vary the cutoff as prescribed. Visualizing the cij with pymol, it looks like (as my test case is an isolated chain from a large complex) there are many very negative correlations, and many positive ones at cij>0.3. So I'm going to actively vary that cutoff.

[rydog]

Also, I can confirm that the ncore=1 options is not required for cna() in my hands.