Mercurial > repos > xuebing > sharplabtool
view tools/taxonomy/poisson2test.py @ 0:9071e359b9a3
Uploaded
author | xuebing |
---|---|
date | Fri, 09 Mar 2012 19:37:19 -0500 |
parents | |
children |
line wrap: on
line source
#!/usr/local/bin/python import sys from math import * from rpy import * if ((len(sys.argv)-1) != 6): print 'too few parameters' print 'usage: inputfile, col1, col2, d-value(not 0), p-val correction method(0 or 1)' sys.exit() try: lines_arr = open(sys.argv[1]).readlines() except IOError: print'cannot open',sys.argv[1] sys.exit() try: i = int(sys.argv[2]) #first column to compare j = int(sys.argv[3]) #second colum to compare d = float(sys.argv[4]) #correction factor k = int(sys.argv[5]) #p-val correction method outfile = open(sys.argv[6],'w') # output data if (i>j): print 'column order not correct col1 < col2' print 'usage: inputfile, col1, col2, d-value, p-val correction method' sys.exit() try: a = 1 / d assert k in [0,1] except ZeroDivisionError: print 'd cannot be 0' print 'usage: inputfile, col1, col2, d-value, p-val correction method' sys.exit() except: print ' p-val correction should be 0 or 1 (0 = "bonferroni", 1 = "fdr")' print 'usage: inputfile, col1, col2, d-value, p-val correction method' sys.exit() except ValueError: print 'parameters are not integers' print 'usage: inputfile, col1, col2, d-value, p-val correction method' sys.exit() fsize = len(lines_arr) z1 = [] z2 = [] pz1 = [] pz2 = [] field = [] if d<1: # Z score calculation for line in lines_arr: line.strip() field = line.split('\t') x = int(field[j-1]) #input column 2 y = int(field[i-1]) #input column 1 if y>x: z1.append(float((y - ((1/d)*x))/sqrt((1/d)*(x + y)))) z2.append(float((2*(sqrt(y+(3/8))-sqrt((1/d)*(x+(3/8)))))/sqrt(1+(1/d)))) else: tmp_var1 = x x = y y = tmp_var1 z1.append(float((y - (d*x))/sqrt(d*(x + y)))) z2.append(float((2*(sqrt(y+(3/8))-sqrt(d*(x+(3/8)))))/sqrt(1+d))) else: #d>1 Z score calculation for line in lines_arr: line.strip() field = line.split('\t') x = int(field[i-1]) #input column 1 y = int(field[j-1]) #input column 2 if y>x: z1.append(float((y - (d*x))/sqrt(d*(x + y)))) z2.append(float((2*(sqrt(y+(3/8))-sqrt(d*(x+(3/8)))))/sqrt(1+d))) else: tmp_var2 = x x = y y = tmp_var2 z1.append(float((y - ((1/d)*x))/sqrt((1/d)*(x + y)))) z2.append(float((2*(sqrt(y+(3/8))-sqrt((1/d)*(x+(3/8)))))/sqrt(1+(1/d)))) # P-value caluculation for z1 and z2 for p in z1: pz1.append(float(r.pnorm(-abs(float(p))))) for q in z2: pz2.append(float(r.pnorm(-abs(float(q))))) # P-value correction for pz1 and pz2 if k == 0: corrz1 = r.p_adjust(pz1,"bonferroni",fsize) corrz2 = r.p_adjust(pz2,"bonferroni",fsize) else: corrz1 = r.p_adjust(pz1,"fdr",fsize) corrz2 = r.p_adjust(pz2,"fdr",fsize) #printing all columns for n in range(fsize): print >> outfile, "%s\t%4.3f\t%4.3f\t%8.6f\t%8.6f\t%8.6f\t%8.6f" %(lines_arr[n].strip(),z1[n],z2[n],pz1[n],pz2[n],corrz1[n],corrz2[n])