Genomic Feature Transformation for Multi-Omics Integration in Clinical Prediction Models

Applying polygenic risk scores (PRS) to multi-omics models enhances disease prediction, outperforming raw genomic data in clinical settings with limited sample sizes.

• Why it matters: Integrating genomics into multi-omics models is challenging due to high dimensionality and sparse information, limiting the effectiveness of current methods and hindering clinical application.
• What they did: Using datasets from AoU and SHCS, the study trained classifiers with four genomic representations—raw SNPs, PCA, PRS, and gene impact scores—and evaluated their performance in predicting CAD and CKD through cross-validation.
• The result: PRS consistently matched or improved prediction accuracy, especially for CAD, and simple feature concatenation was as effective as complex encoders, suggesting additive genomic signals are sufficient for current models.