Package ubic.gemma.core.util.math.linearmodels
package ubic.gemma.core.util.math.linearmodels
-
ClassesClassDescriptionWhat a factor's columns look like under each
ContrastCoding, and what its coefficients therefore mean.Predicting a sample that did not contribute to the fit.Both blocking modes (fixed-effect paired, mixed-model correlation) on a REAL Gemma experiment design: GSE17183 (eid 2877), "Hepatic gene expression before and during interferon and ribavirin combination therapy".Unit tests forLinearModelSummaryUtils.createTerm2CoefficientNames(List, DoubleMatrix): the helper that maps factor names to the set of R-side coefficient-row names belonging to that factor.Situations the golden fixtures do not reach: clamping, missing observations, weights, permutation, and the small-sample gates ofMixedModelFit.estimateCorrelation(DoubleMatrix2D, String[], DoubleMatrix2D, DoubleMatrix2D).The consensus-correlation estimation is a PER-PROBE loop (QR of the design, eigendecomposition of the residual block span, a gamma IRLS) -- the cost scales with probes and could quietly become the new bottleneck of the analysis phase.Mixed model (random intercept): the R reference fits live indata/stat-tests/mixed-golden-gen.R-- limmaduplicateCorrelation(per-probe REML variance decomposition, consensus by trimmed mean on the atanh scale) +gls.series+eBayes, on an UNBALANCED block design: 11 samples over 8 subjects, 3 subjects with both arms, 5 with a single arm.Edge cases of a blocking factor in the design.Paired design: a subject factor as a fixed block, treatment effect adjusted for it.Paired / repeated-measures fits against limma on REAL Gemma designs, one fixture per design shape (seedata/stat-tests/real-designs/fetch_real_designs.pyfor how each was chosen, andreal-designs-golden-gen.Rfor the reference fits; the golden values are read from eachgolden.txt, not copied here).