def omap_to_orths(omap):
orths = []
for i in range(len(omap)):
(organism1, gene1) = omap.keys()[i]
(organism2, gene2) = omap[omap.keys()[i]][0]
orths.append([fr.one_gene(name=gene1,organism=organism1), fr.one_gene(name=gene2, organism=organism2)])
return orths
def eval_network_roc(net, genes, tfs, priors):
from matplotlib import pyplot as plt
org = priors[0][0].organism
priors_set = set(priors)
scores = np.zeros(len(genes)*len(tfs))
labels = np.zeros(len(genes)*len(tfs))
i=0
for tfi in range(len(tfs)):
for gi in range(len(genes)):
tf = tfs[tfi]
g = genes[gi]
score = np.abs(net[tfi, gi])
label = 0
if (fl.one_gene(tf, org), fl.one_gene(g, org)) in priors_set:
label = 1
if (fl.one_gene(g, org), fl.one_gene(tf, org)) in priors_set:
label = 1
scores[i] = score
labels[i] = label
i += 1
(fpr, tpr, t) = roc_curve(labels, scores)
auroc = auc(fpr, tpr)
plt.plot(fpr, tpr)
plt.show()
print auroc
return (labels, scores)
def random_orth(genes1, genes2, organisms, n_orth):
orths = []
for i in range(n_orth):
g1 = genes1[int(np.floor(random.random() * len(genes1)))]
g2 = genes2[int(np.floor(random.random() * len(genes2)))]
orth_group = [fl.one_gene(g1, organisms[0]), fl.one_gene(g2, organisms[1])]
orths.append(orth_group)
return orths
def load_priors(priors_fn, organism):
p = file(priors_fn)
ps = p.read()
psn = filter(len, ps.split('\n'))
psnt = map(lambda x: x.split('\t'), psn)
priors = map(lambda x: (fl.one_gene(x[0], organism), fl.one_gene(x[1], organism)), psnt)
signs = map(lambda x: [-1,1][x[2]=='activation'], psnt)
p.close()
return (priors,signs)
def load_orth(orth_fn, organisms):
f = file(orth_fn)
fs = f.read()
fsn = filter(len, fs.split('\n'))
fsnt = map(lambda x: x.split('\t'), fsn)
#print fsn
orths = []
for o in fsnt:
#print o
orths.append([fl.one_gene(name=o[0],organism=organisms[0]), fl.one_gene(name=o[1], organism=organisms[1])])
return orths
def eval_network_pr(net, genes, tfs, priors):
from matplotlib import pyplot as plt
org = priors[0][0].organism
priors_set = set(priors)
gene_to_ind = {genes[x] : x for x in range(len(genes))}
tf_to_ind = {tfs[x] : x for x in range(len(tfs))}
gene_marked = np.zeros(len(genes)) != 0
tf_marked = np.zeros(len(tfs)) != 0
for prior in priors:
gene_marked[gene_to_ind[prior[0].name]] = True
gene_marked[gene_to_ind[prior[1].name]] = True
if prior[0].name in tf_to_ind:
tf_marked[tf_to_ind[prior[0].name]] = True
if prior[1].name in tf_to_ind:
tf_marked[tf_to_ind[prior[1].name]] = True
genes = np.array(genes)[gene_marked]
tfs = np.array(tfs)[tf_marked]
net = net[:, gene_marked]
net = net[tf_marked, :]
scores = np.zeros(len(genes)*len(tfs))
labels = np.zeros(len(genes)*len(tfs))
i=0
for tfi in range(len(tfs)):
for gi in range(len(genes)):
tf = tfs[tfi]
g = genes[gi]
score = np.abs(net[tfi, gi])
label = 0
if (fl.one_gene(tf, org), fl.one_gene(g, org)) in priors_set:
label = 1
if (fl.one_gene(g, org), fl.one_gene(tf, org)) in priors_set:
label = 1
scores[i] = score
labels[i] = label
i += 1
(precision, recall,t) = precision_recall_curve(labels, scores)#prc(scores, labels)
aupr = auc(recall, precision)
#plt.plot(recall, precision)
#plt.xlabel('recall')
#plt.ylabel('precision')
#plt.title('B subtilis alone, no refitting')
#plt.show()
return aupr
def test_scadBs_refit(b1, b2, fusiongroups, xsamples1, xsamples2, noise_std1, noise_std2, p_falsep, p_falseneg, lamP, lamR, lamS, it, k, s_it):
# TFs = [map(lambda x: str(x)+'tfa', range(b1.shape[0])), map(lambda x: str(x)+'tfb', range(b2.shape[0]))]
# Gs = [map(lambda x: str(x)+'ga', range(b1.shape[1])), map(lambda x: str(x)+'gb', range(b2.shape[1]))]
TFs = [map(lambda x: x, range(b1.shape[0])), map(lambda x: x, range(b2.shape[0]))]
Gs = [map(lambda x: x, range(b1.shape[1])), map(lambda x: x, range(b2.shape[1]))]
xdims1 = (xsamples1, b1.shape[0])
xdims2 = (xsamples2, b2.shape[0])
(x1, y1) = generate_from_linear(xdims1, b1, noise_std1)
(x2, y2) = generate_from_linear(xdims2, b2, noise_std2)
(x1test, y1test) = generate_from_linear(xdims1, b1, noise_std1)
(x2test, y2test) = generate_from_linear(xdims2, b2, noise_std2)
p1 = messwpriors(b1, p_falsep, p_falseneg)
p2 = messwpriors(b2, p_falsep, p_falseneg)
priorset = []
for i in range(len(p1)):
tf = fr.one_gene(TFs[0][p1[i][0]], 0)
gene = fr.one_gene(Gs[0][p1[i][1]], 0)
priorset.append((tf,gene))
for i in range(len(p2)):
tf = fr.one_gene(TFs[1][p2[i][0]], 1)
gene = fr.one_gene(Gs[1][p2[i][1]], 1)
priorset.append((tf,gene))
organisms = [0,1]
# constraints=[]
# for i in range(len(fusiongroups)):
# fg = fusiongroups[i]
# for j in range(len(fg[0])):
# for m in range(len(fg[1])):
# coeff1 = fr.coefficient(0,fg[0][j][0], fg[0][j][1])
# coeff2 = fr.coefficient(1,fg[1][m][0], fg[1][m][1])
# constr = fr.constraint(coeff1,coeff2, lamS)
# constraints.append(constr)
#bs_solve = fr.solve_ortho_scad_refit_bench(organisms, Gs, TFs, [x1, x2], [y1, y2], constraints, priorset, lamP, lamR, lamS, it, k, s_it)
bs_solve = fr.solve_ortho_scad_refit(organisms, Gs, TFs, [x1, x2], [y1, y2], fusiongroups, priorset, lamP, lamR, lamS, it, k, s_it)
err1 = B_err(bs_solve[0], b1)
err2 = B_err(bs_solve[1], b2)
support_score1 = check_support(bs_solve[0], b1)
support_score2 = check_support(bs_solve[1], b2)
err1p = pred_err_grps(bs_solve[0], x1test, y1test)
err2p = pred_err_grps(bs_solve[1], x2test, y2test)
return (bs_solve, err1, err2, err1p, err2p, support_score1, support_score2)
