Introduction

Here, we will prepare the subsetted NHANES totals with males 60-79 years old for PCA and clustering analyses. Unlike ASA24, where we had a choice of using averages across days or treating each day separately, we will have 2-day averages of totals data with NHANES because the NHANES data was collected for only two days.

In this script, we will prepare the totals data by removing variables that have zero variance and collapsing variables by correlation (i.e. removing redundancy of variables that are highly correlated).


Load functions and packages

Name the path to DietR directory where input files are pulled.

main_wd <- "~/GitHub/DietR"

Load the necessary functions.

source("lib/specify_data_dir.R")
source("lib/prep_data_for_clustering.R")

You can come back to the main directory by:

setwd(main_wd)


Load the data and omit some variables if desired

Specify the directory where the data is.

SpecifyDataDirectory(directory.name = "eg_data/NHANES/Laboratory_data/")


Load the totals data that have males 60-79 years old.

totals_males60to79 <- read.table("QCtotal_d_ga_body_meta_glu_comp_2_males60to79.txt", 
                                 sep="\t", header=T)

It should have 236 individuals (rows) and 266 variables (columns).

dim(totals_males60to79)
## [1] 236 266

There may be some variables that you would like to omit before performing PCA.

Define which columns to drop.

drops <- c("KCAL","GRMS", "MOIS", "NoOfItems")

Take only the columns whose names are NOT in the drops vector.

totals_males60to79_2 <- totals_males60to79[ , !(names(totals_males60to79) %in% drops)]


NUT: Nutrients and body weight

Prepare input dataset for clustering with nutrients data and BMI of the participants.

We are interested in BMI because the GLU_index groups had different BMI. Add any other variables of interest that you would like to include in the PCA.

For NHANES, the last nutrient variables is P226, not B12_ADDED as in ASA24.

Obtain the column numbers for start.col=“PROT” through end.col=“P226” plus “BMXBMI” and “SEQN”.

SEQN_col  <- match("SEQN"  , names(totals_males60to79_2)) 
BMI_col   <- match("BMXBMI" , names(totals_males60to79_2)) 
start_col <- match("PROT"   , names(totals_males60to79_2))  
end_col   <- match("P226"   , names(totals_males60to79_2))

Select the BMI, body weight, and the nutrient variables.

user_BMI_nut <- totals_males60to79_2[ , c(SEQN_col, BMI_col, start_col:end_col)]

Process this input, user_BMI_nut, for clustering analysis as follows:

  1. Take complete cases in your variables of interest,
  2. Save the original totals of the complete cases individuals as a .txt,
  3. Keep non-zero columns,
  4. Remove the userID,
  5. Identify correlated variables and remove them,
  6. Save with uncorrelated variables as a .txt,
  7. Save correlation matrix as a .txt.
PrepForClustering(input_df = user_BMI_nut,
                userID = "SEQN",
                original_totals_df= totals_males60to79, 
                complete_cases_fn=   "QCtotal_d_ga_body_meta_glu_comp_2_males60to79_c_Nut.txt",
                clustering_input_fn= "QCtotal_d_ga_body_meta_glu_comp_2_males60to79_c_Nut_rv.txt",
                corr_matrix_fn=      "QCtotal_d_ga_body_meta_glu_comp_2_males60to79_c_Nut_corr_mat.txt")
## user_BMI_nut has 236 rows and 65 variables.
## The following column(s) in user_BMI_nut have missing data shown below.
## Rows (samples) containing those missing data will be removed.
##  BMXBMI 
##      3 
## No columns had zero variance. 
## The numeric data without ID has 233 rows and 64 variables.
## Clustering 64 features...getting means...choosing reps...collapsed from 64 to 38.
## After removing correlated variables, 233 rows and 38 variables remained.


CAT: Food category and body weight

Do the same preparation for the food category data.

Obtain the column numbers for start.col=“F_CITMLB” through end.col=“A_DRINKS” plus “BMXBMI” and “SEQN”.

SEQN_col  <- match("SEQN"     , names(totals_males60to79_2)) 
BMI_col   <- match("BMXBMI"   , names(totals_males60to79_2)) 
start_col <- match("F_CITMLB" , names(totals_males60to79_2))  
end_col   <- match("A_DRINKS" , names(totals_males60to79_2))

Select the BMI, body weight, and the nutrient variables.

user_BMI_cat <- totals_males60to79_2[ , c(SEQN_col, BMI_col, start_col:end_col)]

Prep.

PrepForClustering(input_df = user_BMI_cat,
                  userID = "SEQN",
                  original_totals_df= totals_males60to79, 
                  complete_cases_fn=   "QCtotal_d_ga_body_meta_glu_comp_2_males60to79_c_Cat.txt",
                  clustering_input_fn= "QCtotal_d_ga_body_meta_glu_comp_2_males60to79_c_Cat_rv.txt",
                  corr_matrix_fn=      "QCtotal_d_ga_body_meta_glu_comp_2_males60to79_c_Cat_corr_mat.txt")
## user_BMI_cat has 236 rows and 39 variables.
## The following column(s) in user_BMI_cat have missing data shown below.
## Rows (samples) containing those missing data will be removed.
##  BMXBMI 
##      3 
## No columns had zero variance. 
## The numeric data without ID has 233 rows and 38 variables.
## Clustering 38 features...getting means...choosing reps...collapsed from 38 to 31.
## After removing correlated variables, 233 rows and 31 variables remained.


Come back to the main directory.

setwd(main_wd)