Brain-phenotype models fail for individuals who defy sample stereotypes.

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Individual brain functional organization differences correlate with various traits and behaviors.
• Current linear brain-phenotype models often assume universal applicability, but exhibit variable performance across individuals.
• Understanding model failure is critical for developing robust and unbiased brain-phenotype relationships.

Purpose of the Study:

• To investigate why and in whom predictive models of brain-phenotype relationships fail.
• To identify factors contributing to the failure of generalized brain-phenotype models.
• To propose a framework for improving the accuracy and utility of brain-phenotype models.

Main Methods:

• Utilized predictive modeling, training and testing on independent datasets to ensure generalizability.
• Applied data-driven approach to neurocognitive measures in a heterogeneous dataset, with replication in two external datasets.
• Analyzed model failure patterns in relation to sociodemographic and clinical covariates.

Main Results:

• Brain-phenotype models reflect stereotypical profiles, not unitary cognitive constructs.
• Model failure is associated with individuals whose neurocognitive scores deviate from expected sociodemographic and clinical profiles.
• Model failure is reliable, phenotype-specific, and generalizable across different datasets.

Conclusions:

• A one-size-fits-all approach to brain-phenotype modeling is limited by biased phenotypic measures.
• Stereotypical profiles embedded in models lead to failure when applied to diverse individuals.
• A new framework is proposed to improve model interpretability and identify individualized neural targets for clinical intervention.