MetaGeno: a chromosome-wise multi-task genomic framework for ischaemic stroke risk prediction.

Area of Science:

• Genomics
• Computational Biology
• Cardiovascular Disease Research

Background:

• Genome-wide association studies (GWAS) identify genetic risk factors for ischaemic stroke (IS).
• Current polygenic risk scores (PRS) have limitations in capturing complex genetic interactions.
• Deep neural networks show promise but face challenges with multifactorial diseases like IS.

Purpose of the Study:

• To develop an advanced genomic prediction framework for ischaemic stroke (IS).
• To overcome limitations of traditional PRS by modeling nonlinear genetic interactions.
• To integrate multi-disease learning for enhanced predictive accuracy.

Main Methods:

• Proposed a Chromosome-wise Multi-task Genomic (MetaGeno) framework.
• Utilized a chromosome-based embedding layer for variant interaction modeling.
• Incorporated multi-disease learning and Transformer models for prediction.
• Validated on UK Biobank and All of Us datasets.

Main Results:

• MetaGeno achieved an AUROC of 0.809 on the UK Biobank dataset, outperforming PRS baselines.
• Identified a two-fold increased IS risk in the top 1% risk group.
• Observed a nearly five-fold increase in risk for individuals with modifiable risk factors.
• Demonstrated generalization with an AUROC of 0.764 on the All of Us dataset.

Conclusions:

• The MetaGeno framework offers improved IS genetic risk prediction.
• It effectively identifies high-risk individuals for targeted prevention strategies.
• The model shows potential as a clinical decision-support tool, considering population diversity.