def R_norm_ma_lowess(exp_obj, pairs,
frac=0.75,
value_method="tc"):
"""
Use R lowess.
"""
# Total counts normalized dataframe of counts
if value_method == "tc":
df = norm_tc(exp_obj)
df = utils.drop_null_rows(df, columns=utils.flatten(pairs))
else:
raise Exception("Unknown value method %s" %(value_method))
num_pairs = len(pairs)
print("LOWESS MA normalizing %d pairs" %(num_pairs))
norm_df = df.copy()
norm_df = utils.drop_null_rows(norm_df, columns=utils.flatten(pairs))
for pair in pairs:
x, y = df[pair[0]], df[pair[1]]
M = np.log2(x) - np.log2(y)
# A = average intensity 1/2(XY)
A = 0.5 * (np.log2(x) + np.log2(y))
# Fit M ~ A
corrected_M, corrected_factor = R_run_loess(A, M)
corrected_x = 2**((2*A + corrected_M)/2.)
corrected_y = 2**((2*A - corrected_M)/2.)
del norm_df[pair[0]]
del norm_df[pair[1]]
norm_df[pair[0]] = corrected_x
norm_df[pair[1]] = corrected_y
return norm_df, df
def R_norm_ma_lowess(exp_obj, pairs,
frac=0.75,
value_method="tc"):
"""
Use R lowess.
"""
# Total counts normalized dataframe of counts
if value_method == "tc":
df = norm_tc(exp_obj)
df = utils.drop_null_rows(df, columns=utils.flatten(pairs))
else:
raise Exception, "Unknown value method %s" %(value_method)
num_pairs = len(pairs)
print "LOWESS MA normalizing %d pairs" %(num_pairs)
norm_df = df.copy()
norm_df = utils.drop_null_rows(norm_df, columns=utils.flatten(pairs))
for pair in pairs:
x, y = df[pair[0]], df[pair[1]]
M = np.log2(x) - np.log2(y)
# A = average intensity 1/2(XY)
A = 0.5 * (np.log2(x) + np.log2(y))
# Fit M ~ A
corrected_M, corrected_factor = R_run_loess(A, M)
corrected_x = 2**((2*A + corrected_M)/2.)
corrected_y = 2**((2*A - corrected_M)/2.)
del norm_df[pair[0]]
del norm_df[pair[1]]
norm_df[pair[0]] = corrected_x
norm_df[pair[1]] = corrected_y
return norm_df, df
def norm_ma_lowess(exp_obj, pairs,
frac=0.75,
missing="none",
value_method="tc"):
"""
Run MA-based loess normalization on total
count normalized expression values. Computes
M = log(X/Y)
A = 0.5 * log(X*Y)
where X, Y are total count normalized expression values
for sample X and sample Y, respectively.
Fits loess regression M ~ A and corrects X and Y accordingly.
Assumes input X and Y values are non-logged.
Parameters:
-----------
exp_obj: Experiment object
pairs: Pairs to consider when doing the normalization
frac: frac parameter to lowess()
missing: missing parameter to lowess()
value_method: Method to use to get values for each sample.
If 'tc', then normalize counts by total counts and then use
that for lowess.
Returns a normalized dataframe of values followed by the
dataframe of values used in the normalization (the ones obtained
by 'value_method', since lowess is not done on the counts.)
"""
# Total counts normalized dataframe of counts
if value_method == "tc":
df = norm_tc(exp_obj)
df = utils.drop_null_rows(df, columns=utils.flatten(pairs))
else:
raise Exception("Unknown value method %s" %(value_method))
num_pairs = len(pairs)
print("LOWESS MA normalizing %d pairs" %(num_pairs))
norm_df = df.copy()
norm_df = utils.drop_null_rows(norm_df, columns=utils.flatten(pairs))
for pair in pairs:
x, y = df[pair[0]], df[pair[1]]
M = np.log2(x) - np.log2(y)
# A = average intensity 1/2(XY)
A = 0.5 * (np.log2(x) + np.log2(y))
# Fit M ~ A
corrected_M = run_lowess(A, M,
frac=frac,
missing=missing)
corrected_x = 2**((2*A + corrected_M)/2.)
corrected_y = 2**((2*A - corrected_M)/2.)
del norm_df[pair[0]]
del norm_df[pair[1]]
norm_df[pair[0]] = corrected_x
norm_df[pair[1]] = corrected_y
return norm_df, df
def norm_ma_lowess(exp_obj, pairs,
frac=0.75,
missing="none",
value_method="tc"):
"""
Run MA-based loess normalization on total
count normalized expression values. Computes
M = log(X/Y)
A = 0.5 * log(X*Y)
where X, Y are total count normalized expression values
for sample X and sample Y, respectively.
Fits loess regression M ~ A and corrects X and Y accordingly.
Assumes input X and Y values are non-logged.
Parameters:
-----------
exp_obj: Experiment object
pairs: Pairs to consider when doing the normalization
frac: frac parameter to lowess()
missing: missing parameter to lowess()
value_method: Method to use to get values for each sample.
If 'tc', then normalize counts by total counts and then use
that for lowess.
Returns a normalized dataframe of values followed by the
dataframe of values used in the normalization (the ones obtained
by 'value_method', since lowess is not done on the counts.)
"""
# Total counts normalized dataframe of counts
if value_method == "tc":
df = norm_tc(exp_obj)
df = utils.drop_null_rows(df, columns=utils.flatten(pairs))
else:
raise Exception, "Unknown value method %s" %(value_method)
num_pairs = len(pairs)
print "LOWESS MA normalizing %d pairs" %(num_pairs)
norm_df = df.copy()
norm_df = utils.drop_null_rows(norm_df, columns=utils.flatten(pairs))
for pair in pairs:
x, y = df[pair[0]], df[pair[1]]
M = np.log2(x) - np.log2(y)
# A = average intensity 1/2(XY)
A = 0.5 * (np.log2(x) + np.log2(y))
# Fit M ~ A
corrected_M = run_lowess(A, M,
frac=frac,
missing=missing)
corrected_x = 2**((2*A + corrected_M)/2.)
corrected_y = 2**((2*A - corrected_M)/2.)
del norm_df[pair[0]]
del norm_df[pair[1]]
norm_df[pair[0]] = corrected_x
norm_df[pair[1]] = corrected_y
return norm_df, df
