Integrative analysis of individual-level data and high-dimensional summary statistics.

Area of Science:

• Statistical Genetics
• Bioinformatics
• Computational Biology

Background:

• Statistical analyses typically rely on individual-level data.
• Integrating high-dimensional summary statistics (e.g., from genome-wide association studies) with individual data presents computational challenges.
• Existing methods struggle with numeric issues when optimizing objective functions with many parameters.

Purpose of the Study:

• To develop a procedure for efficient statistical inference by leveraging external summary data.
• To enhance effect estimation and hypothesis testing using combined individual and summary-level data.
• To create a scalable method for high-dimensional genetic data integration.

Main Methods:

• A divide-and-conquer strategy is proposed to handle high-dimensional summary data by breaking down the task into parallel jobs.
• Each job integrates individual-level data with a subset of summary data.
• Final parameter estimates are obtained by pooling results from fitted models using minimum distance estimation.

Main Results:

• The procedure improves the fitting of targeted statistical models, enhancing inference efficiency.
• The method is scalable to high-dimensional summary data.
• Demonstrated advantages through simulations and an application to pancreatic cancer risk prediction using polygenic risk scores.

Conclusions:

• The developed procedure offers a robust and efficient approach for integrating individual-level and high-dimensional summary data.
• The method is applicable to general additive models in genetic studies and can integrate data from different populations.
• An R package (MetaGIM) is available for implementation.