Generalizable predictive modeling of semantic processing ability from functional brain connectivity.

Area of Science:

• Cognitive Neuroscience
• Neuroimaging
• Individual Differences

Background:

• Semantic processing (SP) is crucial for understanding meaningful information.
• Existing neuroimaging studies often average data, overlooking individual variations in neural organization and SP behavior.
• The link between individual differences in neural network organization and SP behavioral variability remains unclear.

Purpose of the Study:

• To identify neural signatures of individual differences in semantic processing (SP).
• To analyze functional connectivity (FC) patterns using predictive modeling on a large dataset.
• To investigate the generalizability of predictive models across different datasets and tasks.

Main Methods:

• Utilized a large-sample Human Connectome Project (HCP) dataset.
• Employed a two-stage predictive modeling approach with internal cross-validation and out-of-sample testing.
• Analyzed functional connectivity (FC) patterns within a putative semantic brain network.

Main Results:

• FC patterns within the semantic network significantly predicted individual SP scores.
• The predictive model generalized to unseen HCP data and partially to independent cohorts.
• Connectivity within the Perisylvian language network was most crucial for prediction and generalization, with better model performance for language tasks and females.

Conclusions:

• Functional connectivity (FC) within specific brain networks underlies individual differences in semantic processing (SP).
• Predictive models based on FC can generalize across datasets and tasks, particularly for language functions.
• Findings support personalized education and interventions for SP and language deficits.