def _ica_fdrtool_signed(E, source, qvalcutoff):
importr("fdrtool")
rfdrtool = ro.r["fdrtool"]
modules = []
print("qvalcutoff: " + str(qvalcutoff))
for source_row in source.T:
rresults = rfdrtool(ro.FloatVector(source_row),
plot=False,
cutoff_method="fndr",
verbose=False)
qvals = np.array(rresults.rx2("qval"))
genes = E.columns[(qvals < qvalcutoff)
& (source_row > source_row.mean())]
modules.append(Module(genes))
genes = E.columns[(qvals < qvalcutoff)
& (source_row < source_row.mean())]
modules.append(Module(genes))
return modules
def _ica_perccutoff(E, source, perccutoff):
modules = []
for source_row in source.T:
sortedgenes = E.columns[source_row.argsort()]
genes = sortedgenes[:int(round(len(E.columns) * perccutoff))]
modules.append(Module(genes))
genes = sortedgenes[-int(round(len(E.columns) * perccutoff)):]
modules.append(Module(genes))
return modules
def _ica_zscore(E, source, stdcutoff):
modules = []
for source_row in source.T:
genes = E.columns[source_row < -source_row.std() * stdcutoff].tolist() + E.columns[source_row > +source_row.std() * stdcutoff].tolist()
modules.append(Module(genes))
return modules
def flame(E, knn=10, threshold=-1, threshold2=-3.0, steps=500, **kwargs):
with TemporaryDirectory() as tmpdir:
with open(tmpdir + "/E.csv", "w") as outfile:
outfile.write(str(E.shape[1]) + " " + str(E.shape[0]) + "\n")
standardize(E).T.to_csv(outfile,
index=False,
header=False,
sep=" ")
binary = os.environ["PERSOFTWARELOCATION"] + "/flame/sample"
command = "{binary} {tmpdir}/E.csv {knn} {threshold2} {steps} {threshold}".format(
**locals())
process = sp.Popen(command, shell=True, stdout=sp.PIPE)
out, err = process.communicate()
modules = []
for row in out.decode().split("\n"):
if row.startswith("Cluster") and "outliers" not in row:
gids = row[row.index(":") + 1:].split(",")
if gids[0] != "":
module = Module([E.columns[int(gid)] for gid in gids])
modules.append(module)
return modules
def genomica(E, R, n=100, **kwargs):
E_genomica = pd.DataFrame(np.vstack([np.array([["desc" for i in range(len(E.columns))], ["desc" for i in range(len(E.columns))]]), scale(E)]), index=["desc", "desc"] + E.index.tolist(), columns=E.columns).T
E_genomica.index.name = "genes"
with TemporaryDirectory() as tmpdir:
E_genomica.to_csv(tmpdir + "/E.csv", sep="\t")
R_genomica = {gid:g for gid, g in enumerate(E_genomica.index) if g in R and g in E.columns}
with open(tmpdir + "/regulators.csv", "w") as outfile:
outfile.write("\n".join([str(gid) for gid in R_genomica.keys()]))
# PERSOFTWARELOCATION is the location in which the software is installed
genomica_loc = os.environ["PERSOFTWARELOCATION"] + "/Genomica/new/"
genomica_command = "cd {genomica_loc};java -XX:ParallelGCThreads=1 -Xmx12G -cp .:../Genomica.jar ExampleProgram ".format(**locals())
n = int(n)
command = genomica_command + "{tmpdir}/E.csv {tmpdir}/regulators.csv {n} 10 > {tmpdir}/output.txt".format(**locals())
sp.call(command, shell=True)
# postprocess output file
state = "members"
modules = []
modulenet = []
with open(tmpdir + "/output.txt") as infile:
for line in infile.readlines():
line = line.rstrip()
if line.startswith("Module members: "):
moduleregulator_scores = defaultdict(float)
state = "members"
module = Module([E.columns[int(geneid)] for geneid in line[len("Module members: "):].split(" ")])
print("----")
print(module)
print(len(modules))
elif line.startswith("<<<<<PROGRAM"):
state = "program"
elif line.find("Regulator: ") > -1:
line = line.split(",")
regulatorid = int(line[0].split(" ")[-1])
#print(line[0].split(" ")[-1])
#print(R_genomica[regulatorid], regulatorid, float(line[-1].split(" ")[-1]))
moduleregulator_scores[R_genomica[regulatorid]] += float(line[-1].split(" ")[-1])
elif line.startswith(">>>>>MODULE"):
#print(moduleregulator_scores)
modules.append(module)
modulenet.append(moduleregulator_scores)
modulenet = pd.DataFrame(modulenet, columns=R).fillna(0)
return modules, modulenet
def _ica_tail(E, source, tailcutoff):
cutoff = np.percentile(source.flatten(), (1-tailcutoff)*100)
modules = []
for source_row in source.T:
genes = E.columns[source_row > cutoff].tolist()
if len(genes) > 0:
modules.append(Module(genes))
return modules
def _ica_fdrtool(E, source, qvalcutoff):
# load fdr
importr("fdrtool")
rfdrtool = ro.r["fdrtool"]
modules = []
for source_row in source.T:
rresults = rfdrtool(ro.FloatVector(source_row), plot=False, cutoff_method="fndr", verbose=False)
qvals = np.array(rresults.rx2("qval"))
genes = E.columns[qvals < qvalcutoff]
modules.append(Module(genes))
return modules
def cmeans(E, k=100, m="auto", cutoff=0.5, cluster_all=True, **kwargs):
importr("Mfuzz")
importr("Biobase")
Exprs = ro.r["ExpressionSet"](ro.r["as.matrix"](standardize(E).T))
if m == "auto":
m = ro.r["mestimate"](Exprs)
rresults = ro.r["mfuzz"](Exprs, k, m)
membership = np.array(rresults.rx2("membership"))
modules = []
for membership_cluster in membership.T:
genes = E.columns[membership_cluster >= cutoff]
modules.append(Module(genes))
return modules