Class BaselineSelectionTest

java.lang.Object
ubic.gemma.core.analysis.expression.diff.BaselineSelectionTest
public class BaselineSelectionTest extends Object

  • Constructor Details

    • BaselineSelectionTest

      public BaselineSelectionTest()

  • Method Details

    • testGetBaselineConditionsRefusesAFactorTheSamplesDoNotCarry

      @Test public void testGetBaselineConditionsRefusesAFactorTheSamplesDoNotCarry()
      A factor none of the given samples carries has no baseline to pick, and getBaselineConditions says so by throwing. The caller therefore has to hand it the factors the samples actually use -- LinearModelAnalyzer's subset path computes exactly that list with fixFactorsForSubset and used to ask for baselines with the unfiltered one. On GSE198008.1 that killed the run: `treatment` applies to two of the four `collection of material` subsets (32/24 and 11/11) and to neither of the other two (15/0 and 15/0).

    • testBaseline

      @Test public void testBaseline()

    • testForcedBaseline

      @Test public void testForcedBaseline()

    • testTermsDiscouragedByTheCurationGuideAreStillDetected

      @Test public void testTermsDiscouragedByTheCurationGuideAreStillDetected()
      The curation guide's "do not use, Gemma won't pick them up" list. Detection is deliberately wider than the guide: these plainly say "this is the control", so they get picked up rather than losing the baseline to an arbitrary choice.

    • testExplicitIsBaselineAlwaysDecides

      @Test public void testExplicitIsBaselineAlwaysDecides()
      An explicit isBaseline decides on its own, whatever else the factor value carries.

    • testMarkedFactorValueWinsOverATermMatch

      @Test public void testMarkedFactorValueWinsOverATermMatch()
      A marked factor value wins the baseline for its factor even when another value matches a control term, regardless of which of the two the factor happens to iterate first.

    • testGetExplicitBaselinesReturnsEveryMarkedValue

      @Test public void testGetExplicitBaselinesReturnsEveryMarkedValue()
      More than one explicitly marked baseline on a factor is legitimate — a dataset holding two experiments has a reference level per experiment. getExplicitBaselines is what lets the analyzer notice that and demand a subset instead of silently taking whichever one it reached first.

    • testGetExplicitBaselinesIgnoresInferredOnes

      @Test public void testGetExplicitBaselinesIgnoresInferredOnes()
      🛑 Only the explicit flag counts. The inference finds two candidates on a great many factors — any design with both a "control" and an "untreated" level — and treating those as an error would fail analyses that run correctly today. Two INFERRED baselines is ambiguity to resolve; two MARKED ones is a curator's statement.