def step4(model_list):
print "Finding the penalty with the best cross-validation likelihood..."
p_list = ["0.05","0.10","0.15","0.20","0.25","0.30","0.35","0.40","0.45","0.50",
"0.55","0.60","0.65","0.70","0.75","0.80","0.85","0.90","0.95","1.0"]
model_name = "-".join(model_list)
model = "+".join(model_list)
for p in p_list:
job_file = job_dir+job_prefix+model_name+"-"+p+".sh"
fout=open(job_file,'w')
if "distance_tss" in model_list:
temp_list = list(model_list)
temp_list.remove("distance_tss")
model_sub = "+".join(temp_list)
command_list = [fgwas, "-i", input_file, "-cc",
"-dists", "distance_tss"+":"+home_dir+"dist_model",
"-w", model_sub, "-p", p, "-xv", "-print", "-onlyp",
"-o", out_dir+model+"-p"+p]
else:
command_list = [fgwas, "-i", input_file, "-cc", "-w",
model, "-p", p, "-xv", "-print", "-onlyp",
"-o", out_dir+model+"-p"+p]
command = " ".join(command_list)
script='''
#$ -N %s%s
#$ -pe shmem 1
#$ -P mccarthy.prjc
#$ -q long.qc
#$ -e %s%s.error
#$ -o %s%s.out
echo "start time" `date`
%s
echo "end time" `date`
''' % (job_prefix,model_name+"-p"+p, log_dir,job_prefix+model_name+"-p"+p,
log_dir,job_prefix+model_name+"-p"+p, command)
fout.write(script)
fout.close()
call = ["qsub", job_file]
out_path = out_dir+model+"-p"+p+".ridgeparams" # try onlyp for comparison
if os.path.exists(out_path) == False:
sp.check_call(call)
if os.path.exists(out_path) == True and os.stat(out_path).st_size == 0:
sp.check_call(call)
job_list = moniter_rescomp_jobs.get_job_ids(job_prefix)
moniter_rescomp_jobs.wait_for_jobs(job_list)
print "Finding best parameter value..."
track_dic = {}
for p in p_list:
fin = open(out_dir+model+"-p"+p+".ridgeparams",'r')
line_list = fin.readlines()
fin.close()
line = line_list[-1]
llk = line.strip().split()[-1]
track_dic[p]=llk
print (p + ": " + str(llk))
sorted_p = sorted(track_dic.items(),key=operator.itemgetter(1))
sorted_p.reverse() #used when comparing ln(lk)
best = sorted_p[0]
print "Optimal parameter value evaluated: %s" % best[0]
return best
def step1():
'''
Start index is the first index for an annotation in the file
Here, the start index is 10 (column 11)
'''
fin = gzip.open(input_file, 'rb')
annot_list = fin.readline().strip().split()[start_index:]
fin.close()
for annot in annot_list:
job_file = job_dir + "job_" + annot + ".sh"
fout = open(job_file, 'w')
if annot == "distance_tss":
command_list = [
fgwas, "-i", input_file, "-cc", "-dists",
annot + ":" + home_dir + "dist_model", "-o", out_dir + annot
]
else:
command_list = [
fgwas, "-i", input_file, "-cc", "-w", annot, "-o",
out_dir + annot
]
command = " ".join(command_list)
# removed #$ -V from script
script = '''
#$ -N job_%s
#$ -pe shmem 1
#$ -P mccarthy.prjc
#$ -q short.qc
#$ -e %s%s.error
#$ -o %s%s.out
echo "start time" `date`
%s
echo "end time" `date`
''' % (annot, log_dir, "job_" + annot, log_dir, "job_" + annot,
command)
fout.write(script)
fout.close()
call = ["qsub", job_file]
out_path = out_dir + annot + ".llk"
if os.path.exists(out_path) == False:
sp.check_call(call)
if os.path.exists(out_path) == True and os.stat(out_path).st_size == 0:
sp.check_call(call)
job_list = moniter_rescomp_jobs.get_job_ids("job_")
moniter_rescomp_jobs.wait_for_jobs(job_list)
def step3(top_annot, top_val,sig_list):
annot_list = sig_list
top_annot_list = top_annot.split("+")
annot_list = [x for x in annot_list if x not in top_annot_list]
out_list = [top_annot+"+"+ x for x in annot_list]
run_models(top_annot_list,annot_list)
job_list = moniter_rescomp_jobs.get_job_ids(job_prefix)
moniter_rescomp_jobs.wait_for_jobs(job_list)
track_dic = {}
for name in out_list:
f = out_dir+name+".llk"
fin=open(f,'r')
#fin.readline() #use for AIC
#fin.readline() #use for AIC
l = fin.readline().strip().split() # get 3rd line
fin.close()
if l[0]=="ln(lk):":
#if l[0]=="AIC:":
track_dic[name] = float(l[1])
sorted_annot = sorted(track_dic.items(),key=operator.itemgetter(1))
sorted_annot.reverse() #used when comparing ln(lk)
#print str(top_val) #+ " : " + str(floor(top_val))
## Next line(s) used for comparing ln(lk)
sig_annot_list = [x for x in sorted_annot if float(x[1]) > float(top_val)]
#sig_list = [x for x in sorted_annot if float(x[1]) > ceil(float(top_val))]
## Next line used for comparing AIC, comment out if unecessary
#sig_list = [x for x in sorted_annot if float(x[1]) < floor(float(top_val))]
#sig_list = [x for x in sorted_annot if float(x[1]) < float(top_val)]
#print sig_annot_list
#print len(sig_annot_list)
if len(sig_annot_list) > 0:
return [sig_annot_list[0], len(sig_annot_list)]
else:
return [top_annot, len(sig_annot_list)]
def step2(sig_list):
print("Finding annotation with highest model likelihood..")
annot_list = sig_list
track_dic = {}
for annot in annot_list:
f = out_dir+annot+".llk"
fin=open(f,'r')
## Use next line for ln(lk) directly
l = fin.readline().strip().split()
## Use next 3 line for AIC
##fin.readline()
##fin.readline()
##l = fin.readline().strip().split()
fin.close()
try:
if l[0]=="ln(lk):":
#if l[0]=="AIC:":
track_dic[annot] = float(l[1])
except:
print f
sorted_annot = sorted(track_dic.items(),key=operator.itemgetter(1))
sorted_annot.reverse() # use for ln(lk)
top_annot = sorted_annot[0][0]
top_val = sorted_annot[0][1]
#print "Top annotation: %s ; ln(lk): %f" % (top_annot, top_val)
##print "Top annotation: %s ; AIC: %f" % (top_annot, top_val)
for annot in sorted_annot[1:]:
annot = annot[0]
#print annot
job_file = job_dir+job_prefix+top_annot+"-"+annot+".sh"
fout=open(job_file,'w')
if annot == "distance_tss":
command_list = [fgwas, "-i", input_file, "-cc",
"-dists", annot+":"+home_dir+"dist_model",
"-w", annot, "-o", out_dir+top_annot+"+"+annot]
else:
command_list = [fgwas, "-i", input_file, "-cc", "-w",
top_annot+"+"+annot, "-o", out_dir+top_annot+"+"+annot]
command = " ".join(command_list)
script='''
#$ -N %s%s
#$ -pe shmem 1
#$ -P mccarthy.prjc
#$ -q short.qc
#$ -e %s%s.error
#$ -o %s%s.out
echo "start time" `date`
%s
echo "end time" `date`
''' % (job_prefix,top_annot+"-"+annot, log_dir,job_prefix+top_annot+"-"+annot,
log_dir,job_prefix+top_annot+"-"+annot, command)
fout.write(script)
fout.close()
call = ["qsub", job_file]
out_path = out_dir+top_annot+"+"+annot+".llk"
if os.path.exists(out_path) == False:
sp.check_call(call)
if os.path.exists(out_path) == True and os.stat(out_path).st_size == 0:
sp.check_call(call)
job_list = moniter_rescomp_jobs.get_job_ids(job_prefix)
moniter_rescomp_jobs.wait_for_jobs(job_list)
top = [top_annot,top_val]
return(top)
def step5(model_list,best_p,best_llk,best_dropped_mod="NA",previously_dropped=[]):
print "Test dropping each annotation from the model, using cross-validation likelihood"
print "Keep dropping annotations as long as the cross-validation likelihood keeps increasing"
if len(previously_dropped) > 0:
dropped = "+".join(previously_dropped) + "+"
else:
dropped = ""
for mod in model_list:
keep_list = list(model_list)
dropped_mod = mod
keep_list.remove(mod)
if len(keep_list) <= 15:
qc = "short.qc"
else:
qc = "long.qc"
keep_mods = "+".join(keep_list)
job_file = job_dir+job_prefix+"drop-"+dropped+mod+".sh"
fout=open(job_file,'w')
if "distance_tss" in keep_list:
keep_list.remove("distance_tss")
model_sub = "+".join(keep_list)
command_list = [fgwas, "-i", input_file, "-cc",
"-dists", "distance_tss"+":"+home_dir+"dist_model",
"-w", model_sub, #keep_mods,
"-p", best_p, "-xv", "-print",
"-o", out_dir+"drop-"+dropped+mod]
else:
command_list = [fgwas, "-i", input_file, "-cc",
"-w", keep_mods, "-p", best_p, "-xv", "-print",
"-o", out_dir+"drop-"+dropped+mod]
command = " ".join(command_list)
script='''
#$ -N %sdrop-%s
#$ -pe shmem 1
#$ -P mccarthy.prjc
#$ -q %s
#$ -e %s%s.error
#$ -o %s%s.out
echo "start time" `date`
%s
echo "end time" `date`
''' % (job_prefix,dropped+mod, qc, log_dir,job_prefix+"drop-"+dropped+mod,
log_dir,job_prefix+"drop-"+dropped+mod, command)
fout.write(script)
fout.close()
call = ["qsub", job_file]
out_path = out_dir+"drop-"+dropped+mod+".ridgeparams"
if os.path.exists(out_path) == False:
sp.check_call(call)
if os.path.exists(out_path) == True and os.stat(out_path).st_size == 0:
sp.check_call(call)
job_list = moniter_rescomp_jobs.get_job_ids(job_prefix)
moniter_rescomp_jobs.wait_for_jobs(job_list)
print "The best likelihood value to beat: %s" % str(best_llk)
track_dic = {}
for mod in model_list:
fin = open(out_dir+"drop-"+dropped+mod+".ridgeparams",'r')
line_list = fin.readlines()
fin.close()
line = line_list[-1]
llk = line.strip().split()[-1]
track_dic[mod]=llk
print ("dropped " + mod + ": " + str(llk))
sorted_mods = sorted(track_dic.items(),key=operator.itemgetter(1))
sorted_mods.reverse() #used when comparing ln(lk)
check_list = [x for x in sorted_mods if float(x[1]) > float(best_llk)]
try:
best = check_list[0]
best_dropped_mod = best[0]
best_dropped_llk = best[1]
report_list = list(model_list)
report_list.remove(best_dropped_mod)
print ("Best dropped model: %s" % best_dropped_mod)
print ("Best dropped llk: %s" % best_dropped_llk)
print ("Annotations to keep: %s" % ",".join(report_list))
status_complete = False
best_llk = best_dropped_llk
return best_dropped_mod,best_dropped_llk, report_list, best_llk, status_complete
except:
print ("Dropping models didn't improve cross-validated likelihood")
print ("Keep the current model!")
status_complete = True
return best_dropped_mod,False,model_list, best_llk, status_complete
