def _mixed_effects_comp(self, ser1, ser2):
groups = [
self.user_indices[i.split('_')[0]] for i in ser1.index.tolist()
]
exo = ser1.values
endo = ser2.values
try:
model = mlm.MixedLM(np.array(endo), np.array(exo),
np.array(groups))
results = model.fit()
summary = results.summary()
l_logger.debug("Summary %r" % (summary, ))
if 'x1' in summary.tables[1].index:
coeff = float(summary.tables[1].loc['x1', 'Coef.'])
pv = results.pvalues[0]
else:
l_logger.warning("no result")
#this happens when a vector has 0 entropy
coeff = None
pv = 1.0
except:
coeff = None
pv = 1.0
l_logger.exception("Error computing mixed model")
raise
return (coeff, pv)
def fitmlm(self):
import pandas as pd
from statsmodels.regression import mixed_linear_model
mlm = mixed_linear_model.MixedLM(endog=pd.DataFrame(self.response),
exog=pd.DataFrame(self.fixed),
groups=pd.DataFrame(self.random),
formula='response ~ fixed')
mlmf = mlm.fit()
return mlmf
def summarymlm(self):
import pandas as pd
from statsmodels.regression import mixed_linear_model
mlm = mixed_linear_model.MixedLM(endog=pd.DataFrame(self.response),
exog=pd.DataFrame(self.fixed),
groups=pd.DataFrame(self.random),
formula='response ~ fixed')
mlmf = mlm.fit()
print(" ")
print("The summary of the linear mixed effects model is given below:")
return mlmf.summary()
def plotmlm(self):
import seaborn as sns
sns.set()
import pandas as pd
from statsmodels.regression import mixed_linear_model
mlm = mixed_linear_model.MixedLM(endog=pd.DataFrame(self.response),
exog=pd.DataFrame(self.fixed),
groups=pd.DataFrame(self.random),
formula='response ~ fixed')
mlmf = mlm.fit()
db_plot = pd.DataFrame()
db_plot["residuals"] = mlmf.resid.values
db_plot["fixed"] = fixed
db_plot["predicted"] = mlmf.fittedvalues
sns.lmplot(x="predicted", y="residuals", data=dbplot)
result_pca.plot_scree()
#Question1: identify variables that correlate with GFP expression
#####Multivariate Linear Regression Model, selection of predictor variables see p.385 R.Johnson
#GFP intens mean can be seen as count variable, that can be modeled using possion/negbino link in GLM.
##[Linear mixed effects model] can be performed for the effects of High/Midium/Low LNP dose
######Here starts the question2: Does internuclear distance correlate with GFP expression?(nuc intens mean/compactness v.s. GFP intens)
# import statsmodels.formula.api as smf
# mixed_model = smf.mixedlm("'GFP intens Mean' ~ 'Nuc intens Mean'", df_mix, groups=df_mix["LNP dose"])
import statsmodels.regression.mixed_linear_model as smm
import statsmodels.regression.mixed_linear_model as smm
Mixed_model = smm.MixedLM(endog=df_mix['GFP intens Mean'].to_numpy(),
exog=df_mix['Nuc intens Mean'].to_numpy(),
groups=df_mix['LNP dose'],
missing='drop') #the model
result_lmm = Mixed_model.fit()
result_lmm.summary()
#result_lmm.summary().as_latex()
####With only distance variables
df_mix_distance = df_mix.rename(
columns={
"Cells no border - Distance from GFP bright Mean":
"Distance from GFP bright Mean",
"Cells no border - Distance intens Mean": "Distance intens Mean"
})
df_mix_distance = df_mix_distance.loc[:, [
'LNP dose', 'GFP intens Mean', 'Nuc intens Mean',
'Distance from GFP bright Mean', 'Distance intens Mean'
def model_playground():
df = pd.read_excel(
r'C:\Users\K1774755\Downloads\phd\mmse_rebecca\mmse_synthetic_data_20190919.xlsx',
index_col=None)
df_smi = df[df.patient_diagnosis_super_class == 'smi only']
df_orga = df[df.patient_diagnosis_super_class == 'organic only']
df_smi_orga = df[df.patient_diagnosis_super_class == 'smi+organic']
df_to_use = df_orga
# MODEL 1: basic model (random intercept and fixed slope)
model = Lmer('score_combined ~ score_date_centered + (1|brcid)',
data=df_to_use) # MMSE score by year
model = Lmer('score_combined ~ score_date_centered + gender + (1|brcid)',
data=df_to_use
) # adding age at baseline as covariate (is this correct??)
to_print = print_r_model_output(model.fit())
# MODEL 2: random intercept and random slope
model = Lmer('score_combined ~ (score_date_centered | brcid)',
data=df_to_use) # this removes the intercept?
model = Lmer('score_combined ~ (score_date_centered + gender| brcid)',
data=df_to_use)
model = Lmer(
"score_combined ~ score_date_centered + (1 + score_date_centered | brcid)",
data=df_to_use) # correct one?
model = Lmer(
'score_combined ~ score_date_centered + gender + (1 + score_date_centered | brcid)',
data=df_to_use)
model = Lmer(
'score_combined ~ 1 + score_date_centered + (1|brcid) + (0 + score_date_centered | brcid)',
data=df) # 2 random effects constrained to be uncorrelated
# MODEL 3: basic model but quadratic
model = Lmer(
'score_combined ~ score_date_centered + I(score_date_centered ^2) + (1|brcid)',
data=df_to_use)
print(model.fit())
#######################################################################################
# PYTHON STUFF
# R formula:
r_formula = 'score_combined ~ score_date_centered + age_at_score_baseline + patient_diagnosis_super_class + score_date_centered * age_at_score_baseline + score_date_centered * patient_diagnosis_super_class'
# MODEL 1: python equivalent
model_py = smf.mixedlm("score_combined ~ score_date_centered ",
df_to_use,
groups=df_to_use['brcid'])
result = model_py.fit()
print(model_py.fit().summary())
# random slope and intercept
model_py = smf.mixedlm(r_formula,
df_to_use,
groups=df_to_use['brcid'],
re_formula="~score_date_centered")
model_py = sm.MixedLM.from_formula(r_formula,
df_to_use,
re_formula="score_date_centered ",
groups=df_to_use['brcid'])
# random slope only
model_py = sm.MixedLM.from_formula("score_combined ~ score_date_centered ",
df_to_use,
re_formula="0 + score_date_centered ",
groups=df_to_use['brcid'])
# MODEL 2: python equivalent (??)
vcf = {
"score_date_centered ": "0 + C(score_date_centered )",
"brcid": "0 + C(brcid)"
}
model_py = sm.MixedLM.from_formula("score_combined ~ score_date_centered ",
groups=df_to_use['brcid'],
vc_formula=vcf,
re_formula="0",
data=df_to_use)
print(model_py.fit().summary())
model3 = mlm.MixedLM(
endog=df_to_use['score_combined'], # dependent variable (1D))
exog=df_to_use[['score_date_centered ',
'intercept']], # fixed effect covariates (2D)
exog_re=df_to_use['intercept'], # random effect covariates
groups=df_to_use['brcid']
) # data from different groups are independent
result = model3.fit()
print(result.summary())