#first sequence set includes randomly distributed sites
#second sequence set includes only one site
#experiment is repeated 100 times
random.seed(None)
#write csv header
out_file.write('Post_motif,Post_single_site\n')
#loop experiments
for cnt in range(0, 100):
print "Experiment: ", cnt
#create background sequence set: 100 seqs 283 bp long
set1 = MGlib.random_DNA(283, {'A': 0.3, 'C': 0.2, 'G': 0.2, 'T': 0.3}, 100)
set2 = set1[:]
#compute softmax scores for background sequences in dataset
gscr = MGlib.esfmax_score_seqs(set1, pssm, rpssm)
#compute softmax scores for motif sequences
mscr = MGlib.esfmax_score_seqs(mot.instances, pssm, rpssm)
#get normal distributions for background and motif
n_g = norm(mean(gscr), std(gscr))
n_m = norm(mean(mscr), std(mscr))
#create motif instances
pmot1 = MGlib.sample_motif(mot, 100)
pmot2 = MGlib.sample_motif(mot, 1)
#insert sites in sequences
e = 0
while (e < len(set1)):
#insert random site in first 10 sequences
#all sites are inserted at the first position of the sequence
#get individual sequence posteriors and write them together with the
#score of the site inserted
#repeat 100 times
random.seed(None)
#write csv header
out_file.write('Ins_site_score,Posterior\n')
#loop experiments
for cnt in range(0,100):
print "Experiment: ", cnt
#create background sequence set: 100 seqs 283 bp long
set1 = MGlib.random_DNA(283,{'A': 0.3,'C': 0.2,'G': 0.2,'T': 0.3},100)
#compute softmax scores for background sequences in dataset
gscr = MGlib.esfmax_score_seqs(set1,pssm,rpssm)
#compute softmax scores for motif sequences
mscr = MGlib.esfmax_score_seqs(mot.instances,pssm,rpssm)
#get normal distributions for background and motif
n_g=norm(mean(gscr), std(gscr))
n_m=norm(mean(mscr), std(mscr))
#create motif instances
pmot1 = MGlib.sample_motif(mot,100)
#insert sites in sequences
e=0
while (e<len(set1)):
set1[e] = pmot1[e] + set1[e]
e = e+1
def main():
###############################################################################
#set default parameters
motif_filename = "CsoR.txt" #input file
out_filename = "cog_exp_sym2_c" #prefix for output
verbose = 0 #verbose mode
alpha = 1.0 / 300.0 #mixing ratio for regulated model
rproms = 3.0 #number of regulated promoters [prior]
tproms = 1811.0 #total number of promoters in genome [prior]
# control number of cogs and number of permutations
num_cogs = 10000
neg_cutoff = 9900 # Cog #'s less than this are negative
num_perms = 100
cog_sample_size = 1000
#verbose
if verbose: print "Using: ", motif_filename, " as input"
if verbose: print "Writing to (suffix): ", "[void]" if out_filename==""\
else out_filename
#open file for ouput
try:
out_file = open(
out_filename + str(num_cogs) + "_s" + str(cog_sample_size) + "_p" +
str(num_perms) + ".csv", "w")
except (IOError, OSError) as file_open_exception:
print "*** Something went wrong while opening the output file"
print "*** Error: ", file_open_exception.errno, " - ",\
file_open_exception.strerror
sys.exit()
#compute priors
PR = rproms / tproms #prior probability of regulation
PB = 1.0 - PR #prior probability of non-regulation
PPR = PB / PR #prior probability ratio
# read motif and assign 0.25 pseudocounts to PSWM
# also assign background uniform distribution for the PSSM (default)
mot = MGlib.read_motif(motif_filename)
mot.pseudocounts = 1
mot.background = None
# save the pssm for the motif and the reverse complement
#(so that they are not recalculated everytime we invoke motif.pssm)
pssm = mot.pssm
rpssm = pssm.reverse_complement()
# Save the motif itself as a list of strings for later permuting
motif_sites = []
num_motif_sites = len(mot.instances)
for i in range(num_motif_sites):
motif_sites.append(str(mot.instances[i]))
random.seed(None)
# Create the COGS
all_cogs = []
the_neg_seqs = []
neg_cog_nums = [i for i in range(0, neg_cutoff)]
ran_neg_cog_nums = sample(cog_sample_size, neg_cog_nums, replace=False)
cog_file = open(
"seqs_sym2_" + str(num_cogs) + "_s" + str(cog_sample_size) + "_p" +
str(num_perms) + ".csv", "w")
for i in range(0, num_cogs):
label = get_cog_type(i, neg_cutoff)
#print "Create cog #", i, label
cur_cog = create_COG(label, mot)
all_cogs.append(cur_cog)
if i in ran_neg_cog_nums:
# A negatively regulated cog
for s in cur_cog:
the_neg_seqs.append(s)
cog_file.write("%d,%s\n" % (i, s))
else:
for s in cur_cog:
cog_file.write("%d,%s\n" % (i, s))
cog_file.close()
# compute softmax scores for sampled background sequences
gscr = MGlib.esfmax_score_seqs(the_neg_seqs, pssm, rpssm)
# compute softmax scores for motif sequences
mscr = MGlib.esfmax_score_seqs(mot.instances, pssm, rpssm)
# get normal distributions for background and motif
mean_gscr = mean(gscr)
std_gscr = std(gscr)
n_g = norm(mean_gscr, std_gscr)
mean_mscr = mean(mscr)
std_mscr = std(mscr)
n_m = norm(mean(mscr), std(mscr))
smeans_file = open(
"smeans_stds_sym2_" + str(num_cogs) + "_s" + str(cog_sample_size) +
"_p" + str(num_perms) + ".csv", "w")
smeans_file.write("PSSM n_g,%13.10f,%13.10f\n" % (mean_gscr, std_gscr))
smeans_file.write("PSSM n_m,%13.10f,%13.10f\n" % (mean_mscr, std_mscr))
# Create the permuted pssm and n_m and n_g for the permutation tests
new_pssm_list = []
rnew_pssm_list = []
n_m_perms = []
n_g_perms = []
for j in range(0, num_perms):
#print "\n***************** Create permutation #", j
# permute the columns of the motif
new_mot = sym_permute_motif(motif_sites)
new_pssm = new_mot.pssm #
rnew_pssm = new_pssm.reverse_complement()
new_pssm_list.append(new_pssm)
rnew_pssm_list.append(rnew_pssm)
# compute score for the negative sequences
gscr = MGlib.esfmax_score_seqs(the_neg_seqs, new_pssm, rnew_pssm)
mean_gscr = mean(gscr)
std_gscr = std(gscr)
# compute softmax scores for new motif sequences
mscr = MGlib.esfmax_score_seqs(new_mot.instances, new_pssm, rnew_pssm)
mean_mscr = mean(mscr)
std_mscr = std(mscr)
smeans_file.write("PermPSSM n_g,%13.10f,%13.10f\n" %
(mean_gscr, std_gscr))
smeans_file.write("PermPSSM n_m,%13.10f,%13.10f\n" %
(mean_mscr, std_mscr))
# get normal distributions for background and motif
n_g_temp = norm(mean_gscr, std_gscr)
n_g_perms.append(n_g_temp)
n_m_temp = norm(mean_mscr, std_mscr)
n_m_perms.append(n_m_temp)
smeans_file.close()
# write csv header
out_file.write(
'COG Num,Pos/Neg Regulated,Posterior,LogLikelihood,True Model LL,LL Pval\n'
)
# For each cog, do the posterior calculation and the permutation tests
for i in range(0, num_cogs):
label = get_cog_type(i, neg_cutoff)
#print "Test Cog:", i,label
# The original posterior computation
#compute softmax scores for sequences in dataset
scrs = MGlib.esfmax_score_seqs(all_cogs[i], pssm, rpssm)
#print np.min(scrs[0]), np.max(scrs[0])
# Compute posterior
# get log-likelihoods for sequences in dataset
llrs = MGlib.ll_ratios(scrs, n_g, n_m, alpha)
# get per-sequence posterior for the sequences in dataset
fpost = MGlib.PostP(llrs, PPR, 0)
true_model_ll = compute_log_l(scrs, n_g, n_m, alpha)
#####################################
# Permutation test
log_ls = []
for j in range(0, num_perms):
#print " ... perm test", j
# Compute score and log likelihood for each permutation.
scrs = MGlib.esfmax_score_seqs(all_cogs[i], new_pssm_list[j],
rnew_pssm_list[j])
log_l = compute_log_l(scrs, n_g_perms[j], n_m_perms[j], alpha)
log_ls.append(log_l)
rev_pval = compute_p_val(log_ls, true_model_ll)
pval = 1.0 - rev_pval
out_file.write("%d,%s,%10.7f,%10.7f,%10.7f,%10.7f\n" %
(i, label, fpost, rev_pval, true_model_ll, pval))
out_file.close()
def main():
###############################################################################
# set default parameters
motif_filename = "CsoR.txt" # input file
out_filename = "cog_exp_sym2_c" # prefix for output
verbose = 0 # verbose mode
alpha = 1.0 / 300.0 # mixing ratio for regulated model
rproms = 3.0 # number of regulated promoters [prior]
tproms = 1811.0 # total number of promoters in genome [prior]
# control number of cogs and number of permutations
num_cogs = 10000
neg_cutoff = 9900 # Cog #'s less than this are negative
num_perms = 100
cog_sample_size = 1000
# verbose
if verbose:
print "Using: ", motif_filename, " as input"
if verbose:
print "Writing to (suffix): ", "[void]" if out_filename == "" else out_filename
# open file for ouput
try:
out_file = open(
out_filename + str(num_cogs) + "_s" + str(cog_sample_size) + "_p" + str(num_perms) + ".csv", "w"
)
except (IOError, OSError) as file_open_exception:
print "*** Something went wrong while opening the output file"
print "*** Error: ", file_open_exception.errno, " - ", file_open_exception.strerror
sys.exit()
# compute priors
PR = rproms / tproms # prior probability of regulation
PB = 1.0 - PR # prior probability of non-regulation
PPR = PB / PR # prior probability ratio
# read motif and assign 0.25 pseudocounts to PSWM
# also assign background uniform distribution for the PSSM (default)
mot = MGlib.read_motif(motif_filename)
mot.pseudocounts = 1
mot.background = None
# save the pssm for the motif and the reverse complement
# (so that they are not recalculated everytime we invoke motif.pssm)
pssm = mot.pssm
rpssm = pssm.reverse_complement()
# Save the motif itself as a list of strings for later permuting
motif_sites = []
num_motif_sites = len(mot.instances)
for i in range(num_motif_sites):
motif_sites.append(str(mot.instances[i]))
random.seed(None)
# Create the COGS
all_cogs = []
the_neg_seqs = []
neg_cog_nums = [i for i in range(0, neg_cutoff)]
ran_neg_cog_nums = sample(cog_sample_size, neg_cog_nums, replace=False)
cog_file = open("seqs_sym2_" + str(num_cogs) + "_s" + str(cog_sample_size) + "_p" + str(num_perms) + ".csv", "w")
for i in range(0, num_cogs):
label = get_cog_type(i, neg_cutoff)
# print "Create cog #", i, label
cur_cog = create_COG(label, mot)
all_cogs.append(cur_cog)
if i in ran_neg_cog_nums:
# A negatively regulated cog
for s in cur_cog:
the_neg_seqs.append(s)
cog_file.write("%d,%s\n" % (i, s))
else:
for s in cur_cog:
cog_file.write("%d,%s\n" % (i, s))
cog_file.close()
# compute softmax scores for sampled background sequences
gscr = MGlib.esfmax_score_seqs(the_neg_seqs, pssm, rpssm)
# compute softmax scores for motif sequences
mscr = MGlib.esfmax_score_seqs(mot.instances, pssm, rpssm)
# get normal distributions for background and motif
mean_gscr = mean(gscr)
std_gscr = std(gscr)
n_g = norm(mean_gscr, std_gscr)
mean_mscr = mean(mscr)
std_mscr = std(mscr)
n_m = norm(mean(mscr), std(mscr))
smeans_file = open(
"smeans_stds_sym2_" + str(num_cogs) + "_s" + str(cog_sample_size) + "_p" + str(num_perms) + ".csv", "w"
)
smeans_file.write("PSSM n_g,%13.10f,%13.10f\n" % (mean_gscr, std_gscr))
smeans_file.write("PSSM n_m,%13.10f,%13.10f\n" % (mean_mscr, std_mscr))
# Create the permuted pssm and n_m and n_g for the permutation tests
new_pssm_list = []
rnew_pssm_list = []
n_m_perms = []
n_g_perms = []
for j in range(0, num_perms):
# print "\n***************** Create permutation #", j
# permute the columns of the motif
new_mot = sym_permute_motif(motif_sites)
new_pssm = new_mot.pssm #
rnew_pssm = new_pssm.reverse_complement()
new_pssm_list.append(new_pssm)
rnew_pssm_list.append(rnew_pssm)
# compute score for the negative sequences
gscr = MGlib.esfmax_score_seqs(the_neg_seqs, new_pssm, rnew_pssm)
mean_gscr = mean(gscr)
std_gscr = std(gscr)
# compute softmax scores for new motif sequences
mscr = MGlib.esfmax_score_seqs(new_mot.instances, new_pssm, rnew_pssm)
mean_mscr = mean(mscr)
std_mscr = std(mscr)
smeans_file.write("PermPSSM n_g,%13.10f,%13.10f\n" % (mean_gscr, std_gscr))
smeans_file.write("PermPSSM n_m,%13.10f,%13.10f\n" % (mean_mscr, std_mscr))
# get normal distributions for background and motif
n_g_temp = norm(mean_gscr, std_gscr)
n_g_perms.append(n_g_temp)
n_m_temp = norm(mean_mscr, std_mscr)
n_m_perms.append(n_m_temp)
smeans_file.close()
# write csv header
out_file.write("COG Num,Pos/Neg Regulated,Posterior,LogLikelihood,True Model LL,LL Pval\n")
# For each cog, do the posterior calculation and the permutation tests
for i in range(0, num_cogs):
label = get_cog_type(i, neg_cutoff)
# print "Test Cog:", i,label
# The original posterior computation
# compute softmax scores for sequences in dataset
scrs = MGlib.esfmax_score_seqs(all_cogs[i], pssm, rpssm)
# print np.min(scrs[0]), np.max(scrs[0])
# Compute posterior
# get log-likelihoods for sequences in dataset
llrs = MGlib.ll_ratios(scrs, n_g, n_m, alpha)
# get per-sequence posterior for the sequences in dataset
fpost = MGlib.PostP(llrs, PPR, 0)
true_model_ll = compute_log_l(scrs, n_g, n_m, alpha)
#####################################
# Permutation test
log_ls = []
for j in range(0, num_perms):
# print " ... perm test", j
# Compute score and log likelihood for each permutation.
scrs = MGlib.esfmax_score_seqs(all_cogs[i], new_pssm_list[j], rnew_pssm_list[j])
log_l = compute_log_l(scrs, n_g_perms[j], n_m_perms[j], alpha)
log_ls.append(log_l)
rev_pval = compute_p_val(log_ls, true_model_ll)
pval = 1.0 - rev_pval
out_file.write("%d,%s,%10.7f,%10.7f,%10.7f,%10.7f\n" % (i, label, fpost, rev_pval, true_model_ll, pval))
out_file.close()
