Cross-ancestry information transfer framework improves protein abundance prediction and protein-trait association...

Area of Science:

• Genetics
• Proteomics
• Computational Biology

Background:

• Genetics-informed proteome-wide association studies (PWAS) are crucial for understanding complex disease mechanisms.
• Current PWAS methods rely on ancestry-matched reference panels, which are limited for underrepresented populations.
• This limitation hinders the discovery of disease-related proteins in diverse ancestries.

Purpose of the Study:

• To develop a novel multi-ancestry framework to improve protein prediction accuracy in underrepresented populations.
• To enhance the identification of proteomic associations with complex traits across diverse ancestries.
• To facilitate gene and protein prioritization for functional validation in multi-omics research.

Main Methods:

• Developed a Multi-Ancestry Best-performing Model (MABM) integrating diverse information-sharing strategies.
• Applied MABM to enhance protein prediction performance in cross-validation and external datasets.
• Utilized the Biobank Japan dataset for PWAS and compared MABM with the Lasso model.

Main Results:

• MABM significantly increased protein prediction performance across diverse populations.
• MABM identified three times more significant PWAS associations compared to the Lasso model in the Biobank Japan dataset.
• 47.5% of MABM-specific associations were successfully reproduced in independent East Asian datasets, demonstrating robustness.

Conclusions:

• The MABM framework effectively enhances protein prediction and PWAS discovery in underrepresented populations.
• MABM facilitates the identification of novel trait-relevant protein candidates and validates known associations.
• This approach broadens the applicability of multi-omics research, particularly in underrepresented groups, and aids trait-relevant protein discovery.