Table 1.
The number of SNPs included in the analysis and the tuning parameters that require cross validation, for the MATURA data sets, for the 11 methods.
| Method | Number of SNPs for anti‐TNF cohort | Number of SNPs for MTX cohort | Tuning parameters that require cross validation |
|---|---|---|---|
| Lasso | |||
| Elastic Net | 40,000 | 42,000 | Penalty parameter |
| Ridge | |||
| RF | 9,000 | 9,000 | Number of variables to split |
| SVR | 9,000 | 9,000 | Standard deviation for Gaussian RBF kernel |
| SPLS | 40,000 | 42,000 | Number of components, sparsity parameter |
| GCTA‐GREML | 4,542,024 | 6,291,430 | NA |
| PRSice* | |||
| BSLMM | 40,000 | 42,000 | NA |
| SkyNet | 9,000 | 9,000 | NA |
| LDpred | 340,000 | 370,000 | NA |
Note. The prediction is based on the SNP effects only. For the PBC data set, we use SVM (Support Vector Machine) instead of SVR on account of binary outcome.
*
For PRSice, the LD‐clumping is performed within the software, resulting in ≈ 140,000 SNPs for the anti‐TNF cohort, and ≈ 170, SNPs for the MTX cohort.