Twin normative modelling project

By Johanna Bayer
Published on July 22, 2022

"In this project I am trying to train and test a normative model on a neuroimaging data set containing twin longitudinal data. I want to look at both changes of deviations in z-scores over time and differences in z-scores between twins"

Project definition

Title

Normative modelling on structural neuroimaging data from the Queensland Twin IMaging (QTIM) data set to compare individualized deviations of twins from the norm.

Background

Normative modelling is known from growth charts, where it describes height in form of an individualized z-score/percentile score of deviation from the norm. Similarly, normative models can be used in neuroimaging to describe an individual’s deviation from a the normative variation of a brain measure, for example from cortical thickness.

Tools

The “project template” project will rely on the following technologies:

Data

The data for this project is the Queensland Twin IMaging (QTIM) dataset: a multimodal neuroimaging dataset of young adult twins and siblings (18-30 years, N = 1026), including a subsample of participants, scanned a second time to assess test-retest reliability (N = 78, test-retest interval ~ 3.5 months).

I am going to use the derivatives from FreeSurfer parcellation that can be downloaded from the repository:

  • cortical thickness data (mm)
  • cortical surface area data (mm^2)
  • subcortical volumes data (mm^3)

Deliverables

At the end of this project, we will have:

  • A first run of the PNC toolkit on the twin data.
  • A package to structure the data to be able to load them into toolkit.
  • Tests that are integrated into the package to test the input data.

Results

I was able to run the model on an initial region the data set, for males and females separately. These are preliminary results:

Predictions for the thickness of the right insula cortex (mm^2):

Predictions for women:

Predictions for men:

Progress overview

  • the model was trained using a 2 fold cross validation
  • the model calculates measures of model fit, such as:
    • explained variance
    • standardized root mean square error
    • Pearson’s correlation coefficient

Tools I learned during this project

  • Deeper understanding of the PNC toolkit I have now a better understanding of the Python implementation, and I customized the implementation in the Jupyter book that is included in this repository.
  • Python packaging I wrote my first ever Python package.
  • Python testing I added some tests to the package, which gave me a deeper understanding of unit tests in Python.
Other projects

Here are other good examples of repositories:

Next steps

  • run the model on remaining brian regions
  • look at changes of z-scores over time within individuals
  • look at twin differences

Conclusion and acknowledgement

I thank the brainhack school staff for all their support and hope to be able to join Brainhack school next year again!

See also these similar projects

Introduction to data visualization in Python

In this module, we will introduce the basics of plotting in python with some of most commonly used packages such as matplotlib and seaborn.

Discover this project

Working with MNE-Python and EEG-BIDS

This repo includes a tutorial for Working with MNE-Python and EEG-BIDS.

Discover this project

Using a machine learning model trained on functional connectivity patterns to predict ADHD

This project uses functional magnetic resonance imaging data to study the connectivity of children with Attention Deficit Hyperactivity Disorder …

Discover this project