def generateY(rep=None,
Genes=None,
Cond=None,
TF=None,
coeff=None,
offset_beta=None,
covar_cond=None,
covar_data=None,
offset_data=None,
frac=None,
fracNoise=None):
'''
Function to generate randomly generated responses Y based on generated coefficients beta
:param rep: number of repetitions to draw Y (Genes x Cond)
:param Genes: first dimension of Y - number of genes
:param Cond: second dimension of Y - number of conditions
:param TF: parameter dimension - number of Transcription Factors
:param coeff: coefficients in model (Genes x TF) - motifs
:param offset_beta: constant variation parameter - sigma_beta^2
:param covar_cond: covariance of conditions (Cond x Cond) - V_Cond
:param covar_data: covariance of data (Cond x Cond) - Sigma_C
:param offset_data: constant variation parameter of data - delta
:return: Y | beta - multivariate normal distributed with N( 0, sigma_beta^2 * V \otimes coeff'coeff.T + delta * Sigma \otimes I_G)
'''
# precompute some values
cov_noise = offset_data * covar_data
cov_motif = offset_beta * covar_cond
if isinstance(offset_beta, list):
if offset_beta.__len__() == 1:
offset_beta = offset_beta[0]
#POOLING
Y_Pool = Pool(max(cpu_count() // 2, 1)).map(
partial(genData,
TF=TF,
Genes=Genes,
Cond=Cond,
coeff=coeff,
cov_motif=cov_motif,
cov_noise=cov_noise), range(rep))
# initialize dataframes
Y = pd.DataFrame(columns=range(rep))
beta = pd.DataFrame(columns=range(rep))
offset_beta_df = pd.DataFrame(columns=range(rep))
# run pover all repetitions
for rand in range(Y_Pool.__len__()):
Ybeta = Y_Pool[rand]
keys = Ybeta.keys()
if keys.__len__() != 2:
raise ('Error: Pooling error')
Y[rand] = pd.Series(Ybeta['Y_r'].reshape(-1, order='F'))
beta[rand] = Ybeta['beta_r'].reshape(-1, order='F')
# bisection:
rsqrd = var.varCoeff(Y=Y[rand], A=coeff, b=Ybeta['beta_r'])
offset_beta_old = offset_beta
lim_low = 0
lim_up = 1e+4 * offset_beta_old
idx = 0
abs_diff = False
# if no fraction betweee nsignal and noise is set, set to 0.2
if frac is None:
frac = 0.2
# find right parameters such that signal is $frac
while not tp.checkRange(rsq=rsqrd, frac=frac) and idx < 1000:
if rsqrd < (frac - 0.01):
if abs_diff:
lim_up = 1e+1 * lim_up
lim_low = offset_beta_old
offset_beta_new = (lim_up - offset_beta_old) / 2
else:
if abs_diff:
lim_low = 0
lim_up = offset_beta_old
offset_beta_new = (offset_beta_old - lim_low) / 2
rsqrd_old = rsqrd
cov_motif_new = offset_beta_new * covar_cond
Ybeta = genData(rand,
Cond=Cond,
Genes=Genes,
TF=TF,
coeff=coeff,
cov_motif=cov_motif_new,
cov_noise=cov_noise)
rsqrd = var.varCoeff(Y=Ybeta['Y_r'].reshape(-1, order='F'),
A=coeff,
b=Ybeta['beta_r'])
offset_beta_old = offset_beta_new
abs_diff = abs(rsqrd - rsqrd_old) < 0.01
idx += 1
# update parameter
offset_beta_df[rand] = [offset_beta_old]
# reshape updated gene expression and motif influence
Y[rand] = Ybeta['Y_r'].reshape(-1, order='F')
beta[rand] = Ybeta['beta_r'].reshape(-1, order='F')
# create dictionary of input parameters
params = {
'G': Genes,
'C': Cond,
'TF': TF,
'sigma_beta2': offset_beta_df,
'delta': offset_data,
'V': covar_cond,
'Sigma': covar_data,
'motif_TG': coeff,
'frac': frac,
'fracNoise': fracNoise
}
# retrun Ybp object
return Ybp.Ybetaparam(Y=Y, beta=beta, parameter=params)
