ADNI Mixed Effect Models

library(HLSGUtils)

Mixed Effect Models

wgs_adno_partitioner function create data partitions from ADNI datasets. To run mixed effect model on ADNI data we have three data source:

  • ADNI dataset
  • ADNI WGS dataset
  • PCA of ADNI WGS data set

We also set the variables from each data source in function arguments: - clinical_variables: variables in ADNI dataset - pca_components: column related to PCA components

After fix input data path, we set number of partitions and partitions save directory.

wgs_adno_partitioner(
  partition_number = 300,
  clinical_variables = c("MMSE", "GENDER","AGE", "MMSE.bl", "PTEDUCAT", "PTID", "VISID"),
  pca_components = c("PC1", "PC2", "PC3"),
  wgs_path = "/data/WGScompletedQC_phenoed.raw",
  pca_path = "/data/ADNIprunedpostQCPCA.eigenvec",
  adni_path = "/data/adnimerge.RData",
  partitions_save_path = "/data/ADNI/data_partitions/")

We need a R script that runs on each core to execute LMER modelling in parallel. function_to_Rscript helps to generate an R script from a function in a package or source from the local code. This script is designed to run independently on each core. To create script we need set:

  • script_name: the name of the created script.
  • function_name: the name of a function in a package or the address of a function’s source file.
  • packages: list of packages that are loaded in the script.
  • arguments: includes function input arguments
  • arguments_class: contains a vector of argument types (character, integer, numeric).
function_to_Rscript(
  script_name = "/scripts/Parallel_Modeling.R", 
  function_from_package = "lmer_modeling",
  packages =  c("HLSGUtils"),
  arguments = c("data_path", "simulation_name", "formula", "save_model_path"),
  arguments_class = c("character", "character", "character", "character")
)

lmer_modeling function is written to run LMER model on each data partitions. It needs :

  • data_path: the path of partition data,
  • formula: lmer formula that contains random effect term,
  • simulation_name: is used in the save model file name.
  • save_model_path: The directory of saving model output.

Finally we use parallel_rscripts to run scripts on parallel cores. The main arguments of function are:

  • rscript_path: path to the script that is run concurrently.
  • args: script’s input arguments.
  • free_memory_treshold: upper bound on memory usage percentage
  • free_cpu_treshold: upper bound on CPU percentage
library(HLSGUtils)

# `lmer_modeling` input arguments
partitions_files = list.files("/data/data_partitions/", full.names = T)
save_model_path = "/data/models/"
formula = paste0("'","MMSE~GENDER+AGE+MMSE.bl+PC1+PC2+PC3+PTEDUCAT+copy_number+(1|PTID)+(1|VISID)","'")
simulation_name = "full_model"

parallel_rscripts(
  rscript_path = "/scripts/Parallel_Modeling.R",
  args = list(data_path = partitions_files,
              simulation_name = simulation_name,
              formula = formula,
              save_model_path = save_model_path),
  used_memory_treshold = 80,
  used_cpu_treshold = 80,
  sleep_time = 10
)

To aggregate models coefficients use aggregate_coefficients function.

aggregate_coefficients(
  save_model_directory = "/data/models/",
  model_names_pattern = "full_model",
  save_model_path = "/data/aggregated_models/full_model.rds"
)