Meta-analysis approach as a gene selection method in class prediction: does it improve model performance? A case study in acute myeloid leukemia

Table 3 Results of the random effects models

Factors	Coefficient	Confidence interval		σ _0C	Confidence interval		σ _0S	Confidence interval		σ _0M(S)	Confidence interval
Factors	Coefficient	LL	UL	σ _0C	LL	UL	σ _0S	LL	UL	σ _0M(S)	LL	UL
n	0.0005	-0.0005	0.0009	0.0244	0.0165	0.0404	0.0489	0.0289	0.0759	0.000	0.000	0.0039
Δ	-0.1169	-0.2041	-0.0285	0.0245	0.0163	0.0402	0.0359	0.0159	0.0405	0.000	0.000	0.0039
ρ	0.1489	0.0295	0.2636	0.0245	0.0165	0.0405	0.0369	0.0022	0.0579	0.000	0.000	0.0039

Each factor was evaluated individually in the random effects linear regression model. The coefficients were inverse transformed to the original scale of the difference of classification model accuracy between MA- and individual classification approach
Abbreviations: LL lower limit, UL upper limit
Symbols: n: the number of samples in each generated dataset; Δ: the log2 fold change of differentially expressed (DE) genes. ρ: pairwise correlation of DE genes. σ _0C, σ _0S and σ _0M(S) are the standard deviation of the random intercepts with respect to classification model, scenario in the simulation study and the number of studies used for selecting relevant features via meta-analysis approach. See Method section for more details regarding the random effect models

ISSN: 1471-2105