def edge_construct(self): for i in range(self.no_of_genes): if self.no_of_cols - ma.sum(self.mask_array[i].mask()) <8: if self.debug: print 'jump_out level 0\t' + self.genelabels[i] continue #less than 8 valid data spots for j in range(i+1, self.no_of_genes): if self.no_of_cols - ma.sum(self.mask_array[j].mask()) <8: if self.debug: print 'jump_out level 1\t' + self.genelabels[j] continue #less than 8 valid data spots mask_tmp = ma.mask_or(self.mask_array[i].mask(), self.mask_array[j].mask()) #joint mask self.cor_vector = [] #initiliation for k in range(self.no_of_cols): new_mask = ma.mask_or(mask_tmp, self.mask_matrix[k]) #leave k out if self.no_of_cols - ma.sum(new_mask) <7: if self.debug: print 'jump_out level 2\t%s v.s %s at %d'%(self.genelabels[i], self.genelabels[j], k,) continue #less than 6, no correlation v1 = ma.array(self.mask_array[i], mask=new_mask).compressed().tolist() v2 = ma.array(self.mask_array[j], mask=new_mask).compressed().tolist() self.cor_vector.append( r.cor(v1,v2)) if len(self.cor_vector) >0: min_cor = min(self.cor_vector) if min_cor >= 0.6: if self.debug: print 'cor vector of %s v.s. %s: %s'%(self.genelabels[i], self.genelabels[j],self.cor_vector,) self.graph_dict[(self.genelabels[i],self.genelabels[j])] = min_cor
def mask_array_construct(self):
data = with_mode(0, r.read_table)(self.dataset_source, row_names=1)
'''
!Important!
if the dataset_source has too few data, conversion from R to python will be a problem.
The whole data matrix will be converted to a python string matrix.
R's NA is not converted to nan in python.
The problem has been found.
r.as_matrix converts small dataset to character type.
r.matrix won't rig the class type, but it rigs the structure.
The only to sovle this is add a colClasses vector to r.read_table.
such as: colClasses=c('character',rep('double',11))
But you have to know the no_of_cols in advance.
As our dataset is really big, this problem hasn't appeared.
'''
#print r.as_matrix(data)
array = ma.masked_inside(r.as_matrix(data), -1.0e20, 1.0e20)
#all are set to be masked except nan. weird! So have to do a converse.
self.mask_array = ma.array(array, mask=ma.logical_not(ma.getmask(array)))
self.genelabels = r.rownames(data)
self.no_of_genes = len(self.genelabels)
self.no_of_cols = len(array[0])
self.mask_matrix=ma.identity(self.no_of_cols)
del array ,data
def edge_construct(self):
for i in range(self.no_of_genes):
#after preprocessing, theses filters are of no use.
'''
if self.no_of_cols - ma.sum(self.mask_array[i].mask()) <self.gene_cut_off:
if self.debug:
sys.stderr.write( 'jump_out level 0\t' + self.genelabels[i])
continue
#less than 8 valid data spots
'''
for j in range(i+1, self.no_of_genes):
'''
if self.no_of_cols - ma.sum(self.mask_array[j].mask()) <self.gene_cut_off:
if self.debug:
sys.stderr.write(print 'jump_out level 1\t' + self.genelabels[j])
continue
#less than 8 valid data spots
'''
joint_mask = ma.mask_or(self.mask_array[i].mask(), self.mask_array[j].mask()) #joint mask
self.cor_vector = [] #initilization
nn_cor_vector = [] #non-negative version of co_vector
for k in range(self.no_of_cols):
new_mask = ma.mask_or(joint_mask, self.mask_matrix[k]) #leave k out
if self.no_of_cols - ma.sum(new_mask) < self.jk_cor_cut_off:
#if self.debug:
# sys.stderr.write( 'jump_out level 2\t%s v.s %s at %d\n'%(self.genelabels[i], self.genelabels[j], k,))
continue
#less than jk_cor_cut_off, no correlation
v1 = ma.array(self.mask_array[i], mask=new_mask).compressed().tolist()
v2 = ma.array(self.mask_array[j], mask=new_mask).compressed().tolist()
cor = r.cor(v1,v2)
self.cor_vector.append( cor)
nn_cor_vector.append(math.fabs(cor))
if self.no_of_cols-ma.sum(joint_mask) == self.jk_cor_cut_off:
break
#Only jk_cor_cut_off(7) valid quantities shared by two genes.
#All the leave-one-out cor's are same. You can only leave NA out.
if len(self.cor_vector) >0:
min_cor = min(nn_cor_vector) #minimum in the non-negative version of cor_vector
if min_cor >= self.cor_cut_off:
if self.debug:
sys.stderr.write('cor vector of %s v.s. %s: %s\n'%(self.genelabels[i], self.genelabels[j],self.cor_vector,))
self.graph_dict[(self.genelabels[i],self.genelabels[j])] = self.cor_vector[nn_cor_vector.index(min_cor)]
def mask_array_construct(self): data =with_mode(0, r.read_table)(self.dataset_source) ''' !Important! if the dataset_source has too few data, conversion from R to python will be a problem. The whole data matrix will be converted to a python string matrix. R's NA is not converted to nan in python. ''' #print r.as_matrix(data) array = ma.masked_inside(r.as_matrix(data), -1.0e20, 1.0e20) #all are set to be masked except nan. weird! So have to do a converse. self.mask_array = ma.array(array, mask=ma.logical_not(ma.getmask(array))) self.genelabels = r.rownames(data) self.no_of_genes = len(self.genelabels) self.no_of_cols = len(array[0]) self.mask_matrix=ma.identity(self.no_of_cols) del array ,data