1. successful compiling of motif clustering (snap/example/motifcluster/) needs a lot of steps and libraries.
Os: Ubuntu linux 12.04 desktop 64bit
1> Install g++
Sudo get-apt…
2>Install gfortran and fortran77
Sudo get-apt gfortran
Note: fortran77 is optional because gfrotran can complile fortran77 code as well. But you need to modify the ARmake.inc file in the root directory of arpack to specify your complier.
3>Install blas
Download from www.netlib.org/blas/
In the BLAS root directory, run make command, get .a files
Mv blas_LINUX.a libplas.a
Sudo cp libblas.a /usr/lib/
Note: 1. .a is static library while .so is dynamic library
2. After you put a libxxx.a file into /usr/lib/, suffix “-lxxx” can be used while compiling.
4>Install lapack
Download from www.netlib.org/lapack/
In the lapack root directory, run make command, get liblapack.a files
Sudo cp liblapack.a /usr/lib/
5>Install arpack
Download: www.caam.rice.edu/software/ARPACK/
First, modify the ARmake.inc file. Especially “home”, where to generate the lib file, the lib’s name, the complier you want to use (f77 by default) and the directory that make lies in. The default directory of make in the os is different from the default value.
Second, modify the second.f in UTIL if needed. (I haven’t)
Third, run “make lib” in the top directory of arpack to get a libxxxx.a
Fourth, “cp libxxxx.a /usr/lib/”
You are all set.
6>Compile motifcluster
Run “make” in the directory “SNAPHOME/example/motifcluter/” to get executable “motifclutermain”.
7>Run it
./motifclustermain -i: C-elegans-frontal.txt -m:bifan
Note: -i: input file, -m: motif pattern
# look the result
Cat C-elegans-frontal-bifan-cluster.txt
Note: the first group of integers is the cluster of most importance. Can be used to get a group of members that are most interesting.
Python get_names.py C-elegans-frontal-meta.csv
Note: to get the name of the nodes in the cluster
2. about the paper
1> I think the explaination of value vol(S) is very misleading. It should be the sum of degrees of all the nodes that reside in S.
2>the algorithm is to find one partition of the s list, an ordered list got from the normalized Laplacian of the motif adjacency matrix, so that the conductance of the partition is the smallest. The smaller group is the cluster we need.