In agapow/subtypr: Robust and Validated Patient Subtyping for Precision Medicine

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 7,
  fig.height = 5
)

Aim

Many methods require data to be in particular ranges or satisfy certain conditions, e.g, a a low amount of (or no) missing data, normalization of all features. Here we demonstrate subtypr for data preprocessing and preparation, including:

• missing data detection & filtering
• missing data imputation
• feature selection
• normalization & more

library(subtypr)

First, let's explore the data with flag_data(), we use is.na() as a detector but it could be function(p) p < 0 if the data was supposed to be positive only.

flag_data(
  m = sparse_methylation,
  flag_fxn = is.na,
  plot = T
)

As we can see, there is only 19 rows without any missing data so the missing data is not due to few patients or few features (as it is sometime) and therefore we'll have to impute missing data.

With the plot we have an idea of the distribution of our problematic data to decide how much patients or features are below 20% percent of missing data for example. Here we can see that the majority is below 0.2 so that's not a big lost

We want to delete patients and features with more than 20% and 15% of missing data respectively. We'll use flag_data(), setting filter = TRUE and putting the values of the treshold for both patients and features in thresh:

filtered_data <- flag_data(
  m = sparse_methylation, flag_fxn = is.na,
  filter = TRUE, thresh = c(0.2, 0.15), print = F
)

Now, let's impute data with pre_process():

imputed_data <- pre_process(filtered_data, method = "knnImpute", k = 5)

agapow/subtypr documentation built on May 5, 2019, 1:33 a.m.