Hybrid hyperalignment: A single high-dimensional model of shared information embedded in cortical patterns of response and functional connectivity

Abstract

Shared information content is represented across brains in idiosyncratic functional topographies. Hyperalignment addresses these idiosyncrasies by using neural responses to project individuals’ brain data into a common model space while maintaining the geometric relationships between distinct patterns of activity or connectivity. The dimensions of this common model capture functional profiles that are shared across individuals such as cortical response profiles collected during a common time-locked stimulus presentation (e.g. movie viewing) or functional connectivity profiles. Hyperalignment can use either response-based or connectivity-based input data to derive transformations that project individuals’ neural data from anatomical space into the common model space. Previously, only response or connectivity profiles were used in the derivation of these transformations. In this study, we developed a new hyperalignment algorithm, hybrid hyperalignment, that derives transformations based on both response-based and connectivity-based information. We used three different movie-viewing fMRI datasets to test the performance of our new algorithm. Hybrid hyperalignment derives a single common model space that aligns response-based information as well as or better than response hyperalignment while simultaneously aligning connectivity-based information better than connectivity hyperalignment. These results suggest that a single common information space can encode both shared cortical response and functional connectivity profiles across individuals.

The Individualized Neural Tuning Model: Precise and generalizable cartography of functional architecture in individual brains

“…In this work, we used searchlight hyperalignment algorithms (Guntupalli et al, 2016) to create a functional template of brain responses based on the training participants. The template is a common, high-dimensional response space, and its column vectors Procrustes algorithm (which preserves representational geometry) (Busch et al, 2021;Feilong et al, 2021Feilong et al, , 2018Guntupalli et al, 2018Guntupalli et al, , 2016Haxby et al, 2020Haxby et al, , 2001Jiahui et al, 2020), WHA calculates transformations using ensemble regularized regression that allows for individualized representational geometries. WHA also introduces a new way to calculate a template matrix M in a single step that more accurately reflects the central tendency for cortical topography and is not biased towards the topography of a "reference brain".…”

“…Despite these differences, our model worked well for both datasets, suggesting it is robust over differences in scan parameters and other details. Recently many largescale neuroimaging datasets have become openly available (Alexander et al, 2017;Horien et al, 2020;Nastase et al, 2021;Snoek et al, 2021;Taylor et al, 2017), and many have naturalistic movie-viewing sessions similar to our datasets. The synergy between our individualized model of brain function and large-scale neuroimaging datasets offers a great opportunity to study individual differences in brain functional organization and their correlates with various phenotypes.…”

Cross-movie prediction of individualized functional topography

“…Using dense connectivity targets (e.g., using all 18742 vertices on the surface) with anatomically-aligned data usually generates poor functional correspondence across participants (Busch et al, 2021). It is, however, beneficial to include more targets for calculating connectivity patterns after the first iteration of connectivity hyperalignment and repeated iterations to lead to a better solution by gradually aligning the information at finer scales.…”

The Normative Modeling Framework for Computational Psychiatry

“…This is an extra critical consideration when using functional MRI as the input. When using fMRI, you should consider using a hyperalignment algorithm (functional alignment) 68,69 to properly model the functional regions, as the spatial overlap of regions across individuals is not guaranteed with functional areas. Progress in addressing spatial autocorrelation in the context of normative modeling has been made 70 , but modeling spatial autocorrelation is a difficult problem that requires further work.…”