cna: Protein Dynamic Correlation Network Construction and Community Analysis.

Description

This function builds both residue-based and community-based undirected weighted network graphs from an input correlation matrix, as obtained from the functions dccm, dccm.nma, and dccm.enma. Community detection/clustering is performed on the initial residue based network to determine the community organization and network structure of the community based network.

Usage

cna(cij, ...)
  ## S3 method for class 'dccm':
cna(cij, cutoff.cij=0.4, cm=NULL,  vnames=colnames(cij),
      cluster.method="btwn", collapse.method="max", 
      cols=vmd.colors(), minus.log=TRUE, ...)
  ## S3 method for class 'ensmb':
cna(cij, \dots, ncore = NULL)

Arguments

cij

A numeric array with 2 dimensions (nXn) containing atomic correlation values, where "n" is the residue number. The matrix elements should be in between 0 and 1 (atomic correlations). Can be also a set of correlation matrices for ensemble netwo

...

Additional arguments passed to the methods cna.dccm and cna.ensmb.

cutoff.cij

Numeric element specifying the cutoff on cij matrix values. Coupling below cutoff.cij are set to 0.

(optinal) A numeric array with 2 dimensions (nXn) containing binary contact values, where "n" is the residue number. The matrix elements should be 1 if two residues are in contact and 0 if not in contact. See the cmap functio

vnames

A vector of names for each column in the input cij. This will be used for referencing residues in a similar way to residue numbers in later analysis.

cluster.method

A character string specifying the method for community determination. Supported methods are: btwn="Girvan-Newman betweenness" walk="Random walk" greed="Greedy algorithm for modularity optimization"

collapse.method

A single element character vector specifing the cij collapse method, can be one of max, median, mean, or trimmed. By defualt the max method is

cols

A vector of colors assigned to network nodes.

minus.log

Logical, indicating whether -log(abs(cij)) values should be used for network construction.

ncore

Number of CPU cores used to do the calculation. By default, use all available cores.

Value

Returns a list object that includes igraph network and community objects with the following components:
networkAn igraph residue-wise graph object. See below for more details.
communitiesAn igraph residue-wise community object. See below for more details.
communitiy.networkAn igraph community-wise graph object. See below for more details.
community.cijNumeric square matrix containing the absolute values of the atomic correlation input matrix for each community as obtained from cij via application of collapse.method.
cijNumeric square matrix containing the absolute values of the atomic correlation input matrix.

Details

The input to this function should be a correlation matrix as obtained from the dccm, dccm.mean or dccm.nma and related functions. Optionally, a contact map cm may also given as input to filter the correlation matrix resulting in the exclusion of network edges between non-contacting atom pairs (as defined in the contact map).

Internally this function calls the igraph package functions graph.adjacency, edge.betweenness.community, walktrap.community, fastgreedy.community. The first constructs an undirected weighted network graph. The second performs Girvan-Newman style clustering by calculating the edge betweenness of the graph, removing the edge with the highest edge betweenness score, calculates modularity (i.e. the difference between the current graph partition and the partition of a random graph, see Newman and Girvan, Physical Review E (2004), Vol 69, 026113), then recalculating edge betweenness of the edges and again removing the one with the highest score, etc. The returned community partition is the one with the highest overall modularity value. walktrap.community implements the Pons and Latapy algorithm based on the idea that random walks on a graph tend to get "trapped" into densely connected parts of it, i.e. a community. The random walk process is used to determine a distance between nodes. Nodes with low distance values are joined in the same community. fastgreedy.community instead determines the community structure based on the optimization of the modularity. In the starting state each node is isolated and belongs to a separated community. Communities are then joined together (according to the network edges) in pairs and the modularity is calculated. At each step the join resulting in the highest increase of modularity is chosen. This process is repeated until a single community is obtained, then the partitioning with the highest modularity score is selected.

Examples

Run this code

##-- Build a correlation network from NMA results
## Read example PDB
pdb <- read.pdb("4Q21")

## Perform NMA
modes <- nma(pdb)
#plot(modes, sse=pdb)

## Calculate correlations 
cij <- dccm(modes)
#plot(cij, sse=pdb)

## Build, and betweenness cluster, a network graph
net <- cna(cij, cutoff.cij=0.35)
#plot(net, pdb)

## within VMD set 'coloring method' to 'Chain' and 'Drawing method' to Tube
#view.cna(net, trim.pdb(pdb, atom.select(pdb,"calpha")), launch=TRUE )

##-- Build a correlation network from MD results
## Read example trajectory file
trtfile <- system.file("examples/hivp.dcd", package="bio3d")
trj <- read.dcd(trtfile)

## Read the starting PDB file to determine atom correspondence
pdbfile <- system.file("examples/hivp.pdb", package="bio3d")
pdb <- read.pdb(pdbfile)

## select residues 24 to 27 and 85 to 90 in both chains
inds <- atom.select(pdb,"///24:27,85:90///CA/")

## lsq fit of trj on pdb
xyz <- fit.xyz(pdb$xyz, trj, fixed.inds=inds$xyz, mobile.inds=inds$xyz)

## calculate dynamical cross-correlation matrix
cij <- dccm(xyz)

## Build, and betweenness cluster, a network graph
net <- cna(cij)

# Plot coarse grained network based on dynamically coupled communities
xy <- plot.cna(net)
plot.dccm2(cij, margin.segments=net$communities$membership)


##-- Begin to examine network structure - see CNA vignette for more details
net
summary(net)
attributes(net)
table( net$communities$members )

Run the code above in your browser using DataLab