Hi all,
I'm trying to perform a cluster analysis on my MD simulation using g_cluster. I'm analysing a simulation of a 11 base pair DNA sequence 10 ns long with 5000 frames.
I have no problem using the program, i runs perfectly using the 'gromos' method.
I have 2 questions, one of which is more advice than a proper question.
1. The out.pdb file produced by g_cluster contains one structure for each cluster identified using a specific cutoff. Is the structure chosen for each cluster a real frame (e.g. the closest to the centroid of the cluster) or an average structure? and if it is the average is it a representative structure?
2. Depending on the -cutoff applied I obtain (of course) different clusters. If i read correctly the default cutoff is 0.1 nm but i can't find much in the literature about this applied to DNA. What would it be a good cutoff for this experiment or what parameter should i consider to make my decision?
Thanks in advance,
Daniele
Hi!
1 - It is not an average structure for each cluster. As you said, it is a frame with least RMSD value in the matrix calculation (thus, the 'center' of the cluster).
2 - Do not take 'default' for 'reference'. Choosing the cutoff, so as the clustering algorithm, should take into account many factors, as conformational ensemble or how resolved you want your clusters to be (lower cutoff = more resolution = more clusters). There are no reference values.
My advice would be to try different values and observe the events in the clusters. You might find an intuitive cutoff.
Hope it helps.
Hi!
1 - It is not an average structure for each cluster. As you said, it is a frame with least RMSD value in the matrix calculation (thus, the 'center' of the cluster).
2 - Do not take 'default' for 'reference'. Choosing the cutoff, so as the clustering algorithm, should take into account many factors, as conformational ensemble or how resolved you want your clusters to be (lower cutoff = more resolution = more clusters). There are no reference values.
My advice would be to try different values and observe the events in the clusters. You might find an intuitive cutoff.
Hope it helps.
Thanks Marcelo, much appreciated.
As you suggested I played around with the cutoff value to find the 'optimal' value. I wanted a compromise between precision and size of clusters/number of clusters and i picked a value accordingly.
Thanks again for you reply.
Daniele
- Badges
- Science topic