Table 1.
Prediction accuracies for GAUDI, PRSice and PRS-CSx for the five traits that GAUDI shows higher average R2 compared to PRS-CSx in UKB quantitative serum and urine biomarker trait screening
| Phenotype | GAUDI R2 Mean (sd) | PRSice R2 Mean (sd) | PRS-CSx R2 Mean (sd) |
|---|---|---|---|
| Lipoprotein A | 0.092 (0.051) | 4.6E-4 (3.5E-4) | 3.0E-3 (2.8E-3) |
| Direct Bilirubin | 0.021 (8.9E-3) | 3.0E-3 (4.7E-3) | 1.2E-4 (7.3E-5) |
| Total Bilirubin | 0.038 (0.013) | 4.8E-4 (5.0E-4) | 4.5E-4 (4.2E-4) |
| Apolipoprotein B | 0.050 (0.017) | 0.016 (0.031) | 0.026 (2.9E-3) |
| LDL direct | 0.032 (4.8E-3) | 4.8E-4 (2.3E-4) | 0.030 (4.3E-3) |
We note that GAUDI and PRSice constructed PRS using the same training samples (N ~ 7000), while PRS-CSx additionally leveraged large scale UKB EUR GWAS (N ~ 430,000). The R2 was calculated as the average of the five-fold outer-loop cross-validation R2. Comparison for a complete list of traits is provided in Supplementary Data 2.