UK Biobank: Predict functional connectivity from white matter microstructure in the brain

The spatial correspondence and genetic influence of inter-hemispheric connectivity with white matter microstructure.
Jeroen Mollink, Stephen M. Smith, Lloyd T. Elliott, Michiel Kleinnijenhuis, Marlies Hiemstra, Fidel Alfaro-Almagro, Jonathan Marchini, Anne-Marie van Cappellen van Walsum, Saad Jbabdi, Karla L. Miller.
Nature Neuroscience, 2019. Available here.

Please send feedback to Jeroen Mollink

Functional connectivity is estimated from resting-state fMRI data between homotopic region pairs (i.e., the homologous brain areas in the left and right cerebral hemispheres). ICA spatial maps are estimated with FSL's MELODIC tool to define these homotopic regions. Probabilistic tractography was run between a pair of regions using diffusion MRI data to delineate the white matter tract connecting them. Microstructure metrics (derived from diffusion MRI) were obtained from the white matter tract connecting a homotopic pair. Tract-based spatial statistics (TBSS) was used to project the microstructure metrics onto a white matter skeleton. For a given homotopic pair, a linear regression model was constructed to predict functional connectivity from the microstructure of the white matter skeleton voxels connecting them.

• We provide code to generate an average time-series for each spatial ICA map. The ICA spatial maps are provided here, together with identifiers listing homoptic regions and non-used components. The remainder of the components is regarded as (physiological) noise. Noise components are regressed out from the average timeseries, after which functional connectivity (using partial correlation) is estimated from the cleaned timeseries between homotopic areas.
• To extract local white matter microstructure, the TBSS masks specific to the white matter tract connecting each homotopic pair are given.
• The final code includes the linear regression model that estimates functional connectivity from white matter microstructure. A principal components analysis is first performed on microstructure matrices to reduce their dimensionality. The top (k=30) microstructural principal components are fed into the design matrix of the regression model. Finally, significance of each model is assessed with permutation testing and a multiple comparison correction is applied.

To access subject specific data (e.g., processed resting-state fMRI and diffusion MRI datasets and confound variables), visit http://www.ukbiobank.ac.uk/register-apply for more information.

More information can be found in the README

You can download the processing pipelines and data here.