Data Integration Methods for Phenotype Harmonization in Multi-Cohort Genome-Wide Association Studies With Behavioral...

Area of Science:

• Behavioral Genetics
• Statistical Genetics
• Psychometrics

Background:

• Genome-wide association studies (GWAS) across multiple cohorts increase power but face challenges.
• Heterogeneity from inconsistent phenotype measurement across studies reduces statistical power in meta-analysis.
• Integrative Data Analysis (IDA) offers a solution for joint phenotype modeling.

Purpose of the Study:

• To investigate IDA for harmonizing behavioral phenotypes across diverse cohorts.
• To introduce a bi-factor integration model (BFIM) for joint phenotype modeling.
• To assess the power and bias of BFIM compared to traditional meta-analysis.

Main Methods:

• Developed a bi-factor integration model (BFIM) to create a common phenotype score.
• Utilized a phenotype reference panel with complete data for model calibration.
• Conducted simulation studies and an empirical demonstration using aggression scores.

Main Results:

• Mega-analysis of genetic variant effects within BFIM showed greater power than meta-analysis on cohort-specific sum scores.
• Using BFIM factor scores in meta-analysis was more powerful than sum scores, with minimal bias.
• A phenotype reference panel was essential for achieving power gains.

Conclusions:

• Model-based harmonization using BFIM is a powerful strategy for genetic consortia.
• BFIM effectively harmonizes phenotype data, accounting for study-specific variability.
• This approach provides a template for analyzing complex traits across multiple datasets.