Improving Phenotyping by Brain-Based Clustering of Cognitive Tests That Rely on Similar Networks
Poster No:
1424
Submission Type:
Abstract Submission
Authors:
Anja Samardzija1, Xilin Shen2, Wenjing Luo2, Abigail Greene2, Saloni Mehta1, Fuyuze Tokoglu2, Jagriti Arora1, Scott Woods2, Rachel Katz2, Gerard Sanacora2, Vinod Srihari2, Dustin Scheinost2, Todd Constable1
Institutions:
1Yale University, New Haven, CT, 2Yale School of Medicine, New Haven, CT
First Author:
Co-Author(s):
Introduction:
We predefine the six cognitive construct networks (attention, perception, declarative memory, language, cognitive control, and working memory), and using Connectome-Based Predictive Modeling (CPM) (Finn, 2015; Shen, 2017) quantify how much each brain network varies with performance on 16 cognitive tests. The predictive power reflects the contribution of that brain network, providing a biologically grounded method for understanding the brain systems upon which test performance relies. We tested the hypothesis that tests that similarly load on the brain across the six networks (as determined by similar predictive power vectors) can be combined to yield composite scores with better phenotypic assessments as reflected by higher predictive power. This introduces a brain-network framework to guide the formation of new composite phenotypic test scores. The goal is not to find the best predictive model, but to use predictive modeling to find the best combination of external measures, knowing that each of the measures selected reflects the same brain systems.
Methods:
A transdiagnostic fMRI dataset was gathered at Yale, involving two resting-state and six task-based runs from 227 participants. The Shen268 (Shen, 2013) atlas was applied to the data. For each participant, time series data from pairs of brain regions were correlated, resulting in eight 268×268 connectivity matrices (one for each imaging run). The six cognitive construct networks of interest were defined from the connectivity matrices by mapping node edges identified through NeuroSynth (https://neurosynth.org/). Specifically, construct network nodes were extracted from NeuroSynth, and the Neurosynth nodes were then aligned with the Shen268 map. Within each connectivity matrix, rows corresponding to each network's selected nodes were included in their entirety. Across runs, the size of these cognitive networks varied between 7×268 and 12×268. To predict behavioral measures, separate kernel ridge regression models were applied to each cognitive network across all tests. The connectivity matrices from the resting and task-based runs were concatenated, resulting in matrices ranging from 7×268×8 to 12×268×8. Predictive performance is the correlation between the model predictions and observed behavioral scores.
To find matching patterns of predictive power across the construct networks, we used hierarchical clustering in which the Euclidean distances were used as the distance metric, and the tests with similar weightings across the six networks were clustered into the same group. The hierarchical clustering was performed across the six networks by 16 test score matrices. Setting an arbitrary distance cutoff of 0.3 yielded five clusters of interest. New composite scores were then made by averaging the individual test scores within each cluster. These composite scores were then modeled using CPM. The fMRI data here doesn't change, only the external brain phenotypic measures – thus improved predictive modeling implies better composite test scores across subjects.
To find matching patterns of predictive power across the construct networks, we used hierarchical clustering in which the Euclidean distances were used as the distance metric, and the tests with similar weightings across the six networks were clustered into the same group. The hierarchical clustering was performed across the six networks by 16 test score matrices. Setting an arbitrary distance cutoff of 0.3 yielded five clusters of interest. New composite scores were then made by averaging the individual test scores within each cluster. These composite scores were then modeled using CPM. The fMRI data here doesn't change, only the external brain phenotypic measures – thus improved predictive modeling implies better composite test scores across subjects.
Results:
Fig. 1 displays the hierarchical clustering tree. Across all six networks the predictive power is higher for the composite score formed by combining the measures in Cluster 1 than for individual test scores in Cluster 1 (Fig. 2a). Similar results are shown for Clusters 2-5 (Fig. 2b-e). It is notable that the six networks combined (magenta) have comparable predictive performance to the whole brain (black), supporting the notion that the six networks sufficiently span the relevant cognitive space even though they do not include the whole-brain.
·Figure 1. Hierarchical clustering of the behavioral measures.
·Figure 2. Predictive power of combined cluster vs. individual behaviors. Circled in red are the predictive powers of the clusters.
Conclusions:
We show that by performing hierarchical clustering on brain-construct model weights, we can derive brain-driven composite test scores that better reflect the relative circuits than any individual score as indicated by higher predictive power in the CPM. This approach is general and can be applied with any network definitions and to any external measures that have clear brain correlates.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling 1
Keywords:
Cognition
Computational Neuroscience
FUNCTIONAL MRI
MRI
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Shen, X., Tokoglu, F., Papademetris, X., & Constable, R. T. (2013). Groupwise whole-brain parcellation from resting-state fMRI data for network node identification. Neuroimage, 82, 403-415.
Shen, X., Finn, E. S., Scheinost, D., Rosenberg, M. D., Chun, M. M., Papademetris, X., & Constable, R. T. (2017). Using connectome-based predictive modeling to predict individual behavior from brain connectivity. nature protocols, 12(3), 506-518.