def roc(args):
""" Calculate ROC_AUC and other metrics and optionally plot ROC curve.
"""
pwmfile = args.pwmfile
fg_file = args.sample
bg_file = args.background
outputfile = args.outfile
# Default extension for image
if outputfile and not outputfile.endswith(".png"):
outputfile += ".png"
motifs = dict([(x.id, x) for x in pwmfile_to_motifs(pwmfile)])
ids = []
if args.ids:
ids = args.ids.split(",")
else:
ids = motifs.keys()
fg_total = {}
result = scan(fg_file, [motifs[x] for x in ids], 0.0, 1)
for key,m in result.items():
fg_total[key.id.split("\t")[0]] = [matches[0][1] for matches in m.values()]
bg_total = {}
result = scan(bg_file, [motifs[x] for x in ids], 0.0, 1)
for key,m in result.items():
bg_total[key.id.split("\t")[0]] = [matches[0][1] for matches in m.values()]
plot_x = []
plot_y = []
# Print the metrics
print "Motif\tROC AUC\tMNCP\tEnr. at 5% FDR\tMax enr."
for id in ids:
fg_vals = fg_total[id]
bg_vals = bg_total[id]
(x, y) = ROC_values(fg_vals, bg_vals)
plot_x.append(x)
plot_y.append(y)
auc = ROC_AUC(fg_vals, bg_vals)
mncp = MNCP(fg_vals, bg_vals)
enr_fdr = enr_at_fdr(fg_vals, bg_vals)
max_enr,score = max_enrichment(fg_vals, bg_vals)
print "%s\t%0.3f\t%03f\t%0.2f\t%0.2f" % (id, auc, mncp, enr_fdr, max_enr)
# Plot the ROC curve
if outputfile:
roc_plot(outputfile, plot_x, plot_y, ids=ids)
def threshold(args):
if args.fdr < 0 or args.fdr > 1:
print "Please specify a FDR between 0 and 1"
sys.exit(1)
motifs = pwmfile_to_motifs(args.pwmfile)
result = scan(args.inputfile, motifs, 0.0, 1)
print "Motif\tScore\tCutoff"
for motif in result.keys():
pwm = motif.pwm
scores = []
min_score = motif.pwm_min_score()
scores = [x[0][1] for x in result[motif].values() if len(x) > 0]
if len(scores) > 0:
opt_score = scoreatpercentile(scores, 100 - (100 * args.fdr))
cutoff = (opt_score - min_score) / (motif.pwm_max_score() - min_score)
print "{0}\t{1}\t{2}".format(motif.id, opt_score , cutoff)
else:
sys.stderr.write("Warning: no matches for {0}\n".format(motif.id))
