The ENIGMA Toolbox: multiscale neural contextualization of multisite neuroimaging datasets

q11

The ENIGMA Toolbox: multiscale neural contextualization of multisite neuroimaging datasets To the Editor-Among big-data neuroscience initiatives, the ENIGMA (Enhancing NeuroImaging Genetics through Meta-Analysis) Consortium-a worldwide alliance of over 2,000 scientists diversified into over 50 working groupshas yielded some of the largest studies of the healthy and diseased human brain.
Through harmonized procedures and by sharing site-specific brain metrics (for example, cortical thickness) or aggregated statistical maps, ENIGMA has set the stage for large-scale analyses comparing findings across different topics or disorders 1,2 . In parallel, increasingly available resources offer opportunities to contextualize findings across multiscale brain organization. Examples include the Allen Human Brain Atlas 3 (microarray-derived postmortem gene expression), the BigBrain Project 4 (three-dimensional postmortem human brain histology) and the Human Connectome Project 5 (high-definition in vivo functional and structural  Over 100 summary statistics from published studies are accessible within the ENIGMA Toolbox. 22q, 22q11.2 deletion syndrome; ADHD, attention deficit hyperactivity disorder; BPD, bipolar disorder; CNVs, copy number variants; DTI, diffusion tensor imaging; EEG, electroencephalography; EOP, early-onset psychosis; GWAS, genome-wide association studies; MDD, major depressive disorder; OCD, obsessive-compulsive disorder; PTSD, post-traumatic stress disorder; rsfMRI, resting-state functional MRI; TBI, traumatic brain injury. b, Surface visualization tools are provided to project cortical and subcortical data results to the surface. As an example, we display gray matter atrophy in left focal epilepsy. c, To contextualize neuroimaging data with respect to microscale brain organization, Toolbox users can fetch disease-related gene expression data (here, we display the average expression levels of focal epilepsy genes; left), stratify atrophy (or other effect maps) according to BigBrain statistical moments (middle left) and gradient (middle right), and stratify atrophy patterns according to cytoarchitectonic classes (right). d, To contextualize neuroimaging data with respect to macroscale brain organization, Toolbox users can load preprocessed functional and structural connectivity data (left), build hub susceptibility models to assess relationships between hub regions and atrophy patterns (middle left) and assess statistical significance of two surface maps using spin permutation testing (middle right), and identify cortical and subcortical disease epicenters that herald patterns of atrophy (right). To advance and simplify cross-disorder analysis and multiscale neural contextualization of neuroimaging findings, the ENIGMA Toolbox offers the ability to access over 100 ENIGMA-derived statistical maps (Fig. 1a), to visualize and manipulate cortical and subcortical surface data and generate publication-ready figures (Fig. 1b) and to contextualize neuroimaging findings at the microscale (postmortem gene expression and cytoarchitecture; Fig. 1c) and macroscale (structural and functional connectome properties; Fig. 1d). To increase generalizability and usability, our Toolbox is compatible with most neuroimaging data and supports the mapping between parcellation maps and vertexwise surface space. To ensure cross-software compatibility, Toolbox users can also export data results to a range of file formats.
As of release v1.1.0, our Toolbox includes 100+ case-control summary statistics from several ENIGMA Working Groups, including 22.q11.2 deletion syndrome, attention deficit hyperactivity disorder, autism spectrum disorder, bipolar disorder, epilepsy, major depressive disorder, obsessive-compulsive disorder and schizophrenia. These datasets, obtained from standardized and quality-controlled protocols, represent morphological (for example, subcortical volume, cortical thickness, surface area) case-control effect sizes from previously published meta-and mega-analyses. Summary statistics from other working groups will be continually added to the Toolbox as they are published.
To deepen our understanding of the molecular and cellular underpinnings of healthy and diseased brain organization, the Toolbox provides microscale neural contextualization workflows to cross-reference neuroimaging findings against the Allen Human Brain Atlas brain-wide gene expression data. Toolbox users can import microarray expression datasets, visualize brain maps of gene expression levels, relate gene expression to properties of brain organization, and identify genes that are spatially correlated with a given brain map. The cell-body-stained BigBrain data 4 and the von Economo and Koskinas cytoarchitectural atlas 6 are also accessible within the Toolbox.
These digitized brain maps link brain microstructure to functional localization and enable users to speculate on the underlying cytoarchitectural composition of, for instance, structurally abnormal areas in specific diseases. When combined, transcriptomic and cytoarchitectonic decoding can embed neuroimaging findings in a rich neurobiological context and yield insights into the etiology of many brain disorders.
At the macroscopic level, network connectivity offers a vantage point to quantify brain reorganization in diseases that are increasingly being conceptualized as network disorders. The Toolbox provides tools to relate surface maps to normative connectome properties derived from functional and structural Human Connectome Project data. Building on prior neurodegeneration 7,8 , psychiatric 9 and epilepsy 10 research, Toolbox users can build hub susceptibility models to assess the vulnerability of highly connected network hubs to disease-related effects and disease epicenter models to identify regions whose connectivity profiles herald disease-related effects. Combined, these two network models can advance our understanding of how connectome architecture relates to morphological abnormalities across a range of disorders. Network models can be further enriched with microstructural properties to inject multiscale information into cortical and subcortical morphometric findings.
By bridging the gap between established data processing protocols and analytic workflows, we hope that the ENIGMA Toolbox facilitates neuroscientific contextualization of results and cross-consortia initiatives. We are eager for researchers and clinicians to test hypotheses beyond traditional case-control comparisons. We hope that our platform will lead to novel and harmonized analyses in global neuroimaging initiatives.

Code availability
Our Toolbox is available in Python and Matlab and complemented with expandable online documentation (http:// enigma-toolbox.readthedocs.io). Derivative data (for example, summary statistics from published ENIGMA studies) and codes are openly accessible under the terms of the BSD-3-Clause license at http://github.com/ MICA-MNI/ENIGMA. Requests to work on a project with subject-level data can be proposed to the working group via the chairs (http://enigma.ini.usc.edu/). Users seeking help are encouraged to subscribe and post their questions to the ENIGMA Toolbox mailing list at https://groups.google.com/g/ enigma-toolbox. ❐