def analyze(inp_fname, out_fname, plt_fname=None, k=3, thresh_init=1e-2, thresh_conv=1e-4, scaling="identity"):
"""Given cleaned data (output from `clean` function), run k-means
clustering analysis, outputting textual and optionally plot output
:param inp_fname: Input filename of raw data
:param out_fname: Output filename for k-means data
:param plt_fname: (optional) Output filename for k-means plot
:param k: (optional) Number of k-means
:param thresh_init: (optional) Threshold for initiation
:param thresh_conv: (optional) Threshold for convergence
:param scaling: (optional) Identity, zscore, or min-max scaling
:returns: None
"""
scaling_funcs = {
"identity": identity,
"zscore": zscore,
"minmax": minmax,
}
dct = load(inp_fname)
keys = nest(dict.keys, filt(bool), tuple)(dct)
rpkm = nest(part(getitems, dct), pipe(scaling_funcs[scaling]))(keys)
sse = 1
sse_last = 0
while relerr(sse, sse_last) > thresh_init:
sse_last = sse
sse, labels, means = kstep(random.sample(rpkm, k), rpkm)
sse = 1
sse_last = 0
while relerr(sse, sse_last) > thresh_conv:
sse_last = sse
sse, labels, _ = kstep(means, rpkm)
means = mstep(labels, rpkm)
if plt_fname is not None:
plot_kmeans(plt_fname, means, labels, rpkm, dct[None], scaling)
organs = pipe(lambda x: x.index(max(x)))(transpose(means))
index = sorted(range(len(organs)), key=organs.__getitem__)
temp = collections.defaultdict(list)
for i in index:
temp[organs[i]].append(dct[None][i])
op = nest(sorted, "|".join, repr)
with open(out_fname, "w") as f:
f.write(f"ID,NAME,GROUP,ASSIGNMENT,URL\n")
for (id_, name), group in sorted(zip(keys, labels), key=nest(reversed, tuple)):
f.write(
f"{id_},{name},{group},{op(temp[group])},{TEMPLATE_NCBI_GENE.format(id_)}\n")
def mstep(labels, values):
"""Generate new k-means given labeled RPKM values
:param labels: K-mean assignments
:param values: tuples of RPKM values
:returns: New k-means
"""
dct = collections.defaultdict(list)
for i, x in zip(labels, values):
dct[i].append(x)
return pipe(nest(transpose, pipe(mean)))(dct.values())
def sorted_fields(data):
"""Sort the fieldnames by mean RPKM
:param data: Pruned data, tuple of dicts
:returns: tuple of sorted fieldnames
"""
full = nest(get(0), dict.keys, tuple)(data)
keys = nest(set, part(sub, set(VALID_KEYS)), sorted, tuple)(full)
vals = pipe(lambda x: pipe(get(x, iskey=False))(keys))(data)
trans = nest(zip, tuple)(*vals)
ind = nest(len, range)(trans)
op = get(pipe(mean)(trans), iskey=False)
srt = nest(part(sorted, key=op, reverse=True), tuple)(ind)
sorted_fields = VALID_KEYS + pipe(get(keys, iskey=False))(srt)
return sorted_fields
def minmax(itr):
if len(itr) <= 1:
return
lo, hi = min(itr), max(itr)
rng = (hi - lo)
op = nest(part(sub, lo), part(truediv, rng))
return pipe(op)(itr)
def convert(dct):
"""Convert a dict of str values to numeric where possible
:param dct: dict with strs from cleaned data
:returns: dict with numeric data
"""
return dict(zip(dct.keys(), pipe(safenum)(dct.values())))
def clean(inp_fname, out_fname):
"""Given raw data (output from `retrieve` function), remove the unnecessary
information and sort the fieldnames by mean RPKM
:param inp_fname: Input filename of raw data
:param out_fname: Output filename for cleaned data
:returns: tuple of dicts
"""
with open(inp_fname) as f:
return nest(csv.DictReader, pipe(prune),
lambda x: writecsv(x, out_fname, sorted_fields(x)))(f)
def load(fname):
"""Load from cleaned data
:param fname: Filename of cleaned data
:returns: dict containing reorganized data
"""
with open(fname) as f:
return nest(
csv.DictReader,
pipe(convert),
part(todatadict)
)(f)
def todatadict(data):
"""Reorganize tuple of dicts into single dict where keys are defined by ID
and NAME, values are numerical data corresponding to RPKM values
:param data: tuple of dicts from loaded cleaned data
:returns: dict containing reorganized data
"""
full = nest(get(0), dict.keys, tuple)(data)
k = nest(filt(nest(VALID_KEYS.__contains__, lambda x: not x)), tuple)(full)
dct = {
(x["ID"], x["NAME"]): pipe(get(x, iskey=False))(k)
for x in data
}
# remember the keys associated with the RPKM values
dct[None] = k
return dct
def retrieve(inp_fname, out_fname, include=bool):
"""Given an search export from NCBI Gene, retrieve RNA-Seq data on those
GeneIDs, and then write out the data to an output file; the `include`
argument is a function that will filter out particular rows of the input
file
:param inp_fname: Input filename of NCBI Gene search export
:param out_fname: Output filename for writing raw data
:param include: Function to filter out input rows
:returns: tuple of dicts
"""
with open(inp_fname) as f:
ids = nest(part(csv.DictReader, delimiter="\t"), filt(include),
apply(nest(get("GeneID"), int)), sorted, tuple)(f)
urls = pipe(TEMPLATE_NCBI_GENE.format)(ids)
n = (1 + len(urls) // CHUNK_SIZE)
print(f"A total of {n} chunks will be retrieved.")
return procfull(out_fname)(urls)
def zscore(itr):
mu = mean(itr)
s = sd(itr)
op = nest(part(sub, mu), part(truediv, s))
return pipe(op)(itr)
def var(itr):
if len(itr) <= 1:
return float("nan")
mu = mean(itr)
op = nest(part(sub, mu), part(pow, 2))
return sum(pipe(op)(itr)) / (len(itr) - 1)
def harmonic_mean(itr):
return inv(mean(pipe(inv)(itr)))