MBimpute: Model-Based Imputation of missing values

In ProteoMM: Multi-Dataset Model-based Differential Expression Proteomics Analysis Platform

Description

Impute missing values based on information from multiple peptides within a protein Expects the data to be filtered to contain at least one observation per treatment group. For experiments with lower overall abundaneces such as multiplexed experiments check if the imputed value is below 0, if so value is reimputed untill it is above 0.

Usage

MBimpute(mm, treatment, prot.info, pr_ppos = 2, my.pi = 0.05,
  compute_pi = FALSE)

Arguments

mm	number of peptides x number of samples matrix of intensities
treatment	vector indicating the treatment group of each sample eg as.factor(c('CG','CG','CG', 'mCG','mCG','mCG')) or c(1,1,1,1,2,2,2,2)
prot.info	protein metadata, 2+ columns: peptide IDs, protein IDs, etc
pr_ppos	column index for protein ID in prot.info
my.pi	PI value, estimate of the proportion of peptides missign completely at random, as compared to censored at lower abundance levels default values of 0.05 is usually reasoanble for missing completely at random values in proteomics data
compute_pi	TRUE/FALSE (default=FALSE) estimate Pi is set to TRUE, otherwise use the provided value. We consider Pi=0.05 a reasonable estimate for onservations missing completely at random in proteomics experiments. Thus values is set to NOT estimate Pi by default. Note: spline smoothing can sometimes produce values of Pi outside the range of possible values.

Value

A structure with multiple components

y_imputed

number of peptides x m matrix of peptides with no missing data

imp_prot.info

imputed protein info, 2+ columns: peptide ID, protein IDs, etc Dimentions should be the same as passed in

Examples

data(mm_peptides)
head(mm_peptides)
intsCols = 8:13 # different from parameter names as R uses outer name spaces
                # if variable is undefined
metaCols = 1:7 # reusing this variable
m_logInts = make_intencities(mm_peptides, intsCols)  # will reuse the name
m_prot.info = make_meta(mm_peptides, metaCols)
m_logInts = convert_log2(m_logInts)
grps = as.factor(c('CG','CG','CG', 'mCG','mCG','mCG'))

set.seed(135)
mm_m_ints_eig1 = eig_norm1(m=m_logInts,treatment=grps,prot.info=m_prot.info)
mm_m_ints_eig1$h.c # check the number of bias trends detected
mm_m_ints_norm = eig_norm2(rv=mm_m_ints_eig1)
mm_prot.info = mm_m_ints_norm$normalized[,1:7]
mm_norm_m =  mm_m_ints_norm$normalized[,8:13]

# ATTENTION: SET RANDOM NUMBER GENERATOR SEED FOR REPRODUCIBILITY !!
set.seed(125) # if nto set every time results will be different
imp_mm = MBimpute(mm_norm_m, grps, prot.info=mm_prot.info, pr_ppos=2,
                  my.pi=0.05, compute_pi=FALSE)

ProteoMM documentation built on Nov. 8, 2020, 5:57 p.m.