find_DMR: Finding DMR

In geneticsMiNIng/metR: Tools for Identification and Visualisation of Differentially Methylated Regions

Description

Finding DMR by Wilcoxon, t-Student, Kolmogorov-Smirnow tests or logistic regression, logistic regression with mixed models, logistic regression with mixed models with correlation matrix. In Ttest, Wilcoxon and Ks are compared methylation rate between x and y prob on the same position and chromosome and null hypothesis is that mean (median, distribution respectively) of methylation rate. In these methods alternative hypothesis is two sided and these sorts regions based on criticial value.

Usage

find_DMR(data, methods, p.value.log.reg = NULL, p.value.reg.mixed = NULL,
  p.value.reg.corr.mixed = NULL, beta.coef.max = 30)

Arguments

data	There are two options: 1. dataframe with specific columns: chr, poz, prob, no, meth, unmeth, meth.rate. This dataframe is result of function preprocessing. 2. dataframe with specific columns: chr, poz, prob, no, meth, unmeth, meth.rate, tiles and possible tiles.common columns. This dataframe is result of function create.tiles.min.gap or create.tiles.fixed.length.
methods	vectors with given methods. Possible values are: 'Wilcoxon', 'Ttest', 'KS', 'Reg.Log', 'Reg.Mixed', 'Reg.Corr.Mixed'. 'Wilcoxon' - Wilcoxon signed test; 'Ttest' - t-Student test with unequal variance; 'KS' - Kolmogorov-Smirnov test; 'Reg.Log' - Wald test of grouping variable from logistic regression; 'Reg.Mixed' - Wald test of grouping variable from logistic regression with mixed effects; 'Reg.Corr.Mixed' - Wald test of grouping variable from logistic regression with mixed effect and estimated previous correlation matrix
p.value.log.reg	if not NULL regions with p.value of prob variable smaller than p.value.log.reg are returned and decreasingly ordered by absolute value of beta coefficient of prob variable otherwise regions ale increasingly ordered by p.value
p.value.reg.mixed	if not NULL regions with p.value of prob variable smaller than p.value.log.reg are returned and decreasingly ordered by absolute value of beta coefficient of prob variable otherwise regions ale increasingly ordered by p.value
p.value.reg.corr.mixed	if not NULL regions with p.value of prob variable smaller than p.value.log.reg are returned and decreasingly ordered by absolute value of beta coefficient of prob variable otherwise regions ale increasingly ordered by p.value
beta.coef.max	only results which have absolute value of beta.coef less than this parameter are returned from Log.Reg, Reg.Mixed, Reg.Corr.Mixed. This prevent cases when algorithm did not convergence well

Details

In regression methods, number of success are number of methylated citosines and failures are number of unmethylated citosines. Output from this methods is beta coefficient of indicator variable from regression model and criticial value from Wald test on indicator variable. Indicator variable is equal 1 if observations are from x prob and 0 otherwise. These methods order regions based on beta coefficients of grouping variable or p.values of grouping variable. In mixed models explantatory variable is only indicator variable and positions of chromosome are random effects. In standard logistic regression explantatory variables are also position of chromosome.

Value

list object. Elements of list are results of given methods. The most interesting regions are on the top

Examples

data('schizophrenia')
control <- schizophrenia %>% filter(category == 'control') %>%
dplyr::select(-category)

disease <- schizophrenia %>% filter(category == 'disease') %>%
 dplyr::select(-category)

data <- preprocessing(control, disease)
data.tiles <- create_tiles_max_gap(data, gaps.length = 100)
data.tiles.small <- data.tiles %>% filter(tiles < 30)

 #finding DMR by all methods with sorting on p.values
find_DMR(data.tiles.small, c('Wilcoxon', 'Ttest', 'KS', 'Reg.Log', 'Reg.Mixed', 'Reg.Corr.Mixed'))

#finding DMR by 'Reg.Log', 'Reg.Mixed', 'Reg.Corr.Mixed'  with sorting on beta values
find_DMR(data.tiles.small, c('Reg.Log', 'Reg.Mixed', 'Reg.Corr.Mixed'), p.value.log.reg = 0.01, p.value.reg.mixed = 0.02, p.value.reg.corr.mixed=0.03)

#finding DMR only by 'Reg.Log' with sorting on beta values and 'Wilcoxon' with sorting on p.values
find_DMR(data.tiles.small, c('Wilcoxon', 'Reg.Log'), p.value.log.reg = 0.001)

geneticsMiNIng/metR documentation built on May 28, 2019, 8:41 p.m.