From: Single-center versus multi-center data sets for molecular prognostic modeling: a simulation study
Sc1 | Sc2 | Sc3 | |||
---|---|---|---|---|---|
signal strength | \(\tilde {\beta }\) | = | [0; 0.5] | 0.25 | 0.125 |
number of genes, informative | ninf | = | 300 | [1;1000] | 300 |
sample size | Ns | = | 100 | 100 | [40 500] |
number of genes, total | Ng | = | 103 | 103 | 103 |
number of centers in MC | Nc | = | 8 | 8 | 8 |
minimum samples per center | nmin | = | 10 | 10 | 5 |
basal level gene g | αg | ∼ | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) |
target | aij | ∼ | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) |
fixed batch effect gene g | γjg | ∼ | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) |
number of latent factors | mj | = | 5 | 5 | 5 |
factor loadings | bjgl | ∼ | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) |
impact of factor l on sample i | Zijl | ∼ | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) |
noise scaling of gene g in batch j | δjg | ∼ | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) |
noise | εijg | ∼ | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) | \(\mathcal {N}(0,1)\) |
standard deviation of observation noise | σy | = | 0.1 | 0.1 | 0.1 |