Skip to content

Latest commit

 

History

History
98 lines (66 loc) · 5.11 KB

README.md

File metadata and controls

98 lines (66 loc) · 5.11 KB
Raw

abcd_multiview

This repository supports the manuscript titled Bayesian Integrative Mixed Modeling Framework for Analysis of the Adolescent Brain and Cognitive Development Study. The analysis focuses on multiview data from the ABCD Study® to integrate heterogeneous data types, account for nested hierarchical structures, and predict behavioral outcomes using the BIPmixed framework. This framework extends the Bayesian Integrative Analysis and Prediction (BIP) model to incorporate 2-level random effects, improving predictive performance and interpretability for data with hierarchical structures.

Authors

Aidan Neher, Apostolos Stamenos, Mark Fiecas, Sandra Safo, Thierry Chekouo @ Biostatistics and Health Data Science, University of Minnesota

Repository Structure

The repository includes the following:

Top-Level Files:

  • 00_abcd_eda.docx: Initial exploratory data analysis (EDA) document.
  • 00_abcd_eda.qmd: Quarto file for the EDA.
  • 01_load_covars_and_outcomes.R: Script for loading covariates and outcome data.
  • 01_load_ELA.R: Script for processing the Early Life Adversity (ELA) view.
  • 01_load_imaging_data.R: Script for processing imaging views (structural and functional MRI).
  • 02_combine_data_define_sample.R: Combines and merges the views to define the analysis sample.
  • 03_generate_scree_plots.R: Generates scree plots for determining latent component hyperparameter r.
  • data_analysis.R: Main script for performing BIPmixed and BIP data analysis.
  • data_analysis_summary.R: Summarizes the results of the BIPmixed and BIP analysis.
  • simulation_study.R: Script for running simulation studies.
  • simulation_study_summary.R: Summarizes results from simulation studies.

Folders:

  • data/: Contains locally stored data for analysis. Note that this directory is excluded by .gitignore to ensure data privacy, so you will need to populate accordingly.
  • figures/: Stores output visualizations, e.g. scree plots and Sankey diagrams.
  • src/: Contains the core code for the BIPmixed method and supporting functions.
  • data_analysis_results/: Directory for output files related to data analysis results.
  • simulation_study_results/: Directory for output files from simulation studies.

Workflow

  1. Exploratory Data Analysis (EDA):

    • Scripts and documents with the prefix 00_ outline the initial exploration of data.

  2. Data Loading and Processing:

    • Scripts with the prefix 01_ focus on loading and processing the individual views of data:
      • ELA metrics.
      • Structural MRI (cortical thickness and surface area).
      • Functional MRI (network correlations).
      • Covariates and outcomes.

  3. Sample Definition:

    • 02_combine_data_define_sample.R merges processed views into a unified dataset, ensuring consistent samples across views.

  4. Hyperparameter Selection:

    • 03_generate_scree_plots.R calculates eigenvalues of concatenated views and determines the number of latent factors using a scree plot.

  5. BIPmixed Analysis:

    • data_analysis.R applies the BIPmixed framework to the processed data, performing feature selection and outcome modeling.
    • Results are summarized in data_analysis_summary.R.

  6. Simulation Studies:

    • simulation_study.R simulates multiview data under varying random effect scenarios to evaluate BIPmixed and alternative methods.
    • Results are summarized in simulation_study_summary.R.

Key Features

  • Exploratory Data Analysis:

    • Initial data exploration guides decisions about feature inclusion, exclusion, and preprocessing.

  • BIPmixed Framework:

    • Integrates multiview data with hierarchical structures.
    • Simultaneously performs feature selection and outcome prediction.

  • Simulation Studies:

    • Compare BIPmixed against baseline BIP, PCA2Step, and Cooperative Learning methods across different hierarchical scenarios discussed in the manuscript.

Outputs

  • Figures:

    • Where scree plots for latent factor selection and Sankey plots showing feature-to-component mapping are written.

  • Summary Tables:

    • Report feature selection performance (e.g., FPR, FNR, AUC) and prediction accuracy (e.g., MSE), and are written to data_analysis_results or simulation_study_results.

Instructions for Reproduction

To reproduce the analyses or simulations, ensure the following:

  1. Data:

    • Place required data files in the data/ directory.

  2. Execution:

    • Run scripts sequentially according to the workflow described above, installing the required R packages as you go.

Contact

For questions or collaborations, please contact Aidan Neher at [email protected].

The authors gratefully acknowledge the research support provided by the NIH T32 Interdisciplinary Biostatistics Training in Genetics and Genomics program (T32 GM132063, 2020–2025). Thierry Chekouo was supported by a National Institutes of Health (NIH) grant: 1R35GM150537-01. Thierry Chekouo also thanks Medtronic Inc. for their support in the form of a faculty fellowship. Sandra Safo was supported by NIH NIGMS grant award number 1R35GM142695.