Repository navigation
Immutable
release. Only release title and notes can be modified.
⚠️ Changed results
- Mixscape selects markers and classifies cells like Seurat's
RunMixscape: tie-corrected Wilcoxon tests with Bonferroni correction over all features, a 10% detection filter, at least three cells per group and Seurat's fold change on log1p-normalizedde_layerdata, then KO calls from the likelihood ratio of the fitted components without their mixing weights and an EM run to the tolerance ofnormalmixEM; KO/NP agreement with Seurat 5.5.0 on Papalexi 2021 goes from 96.6% to 99.1% (#1197) @kkkklck, @Zethson
🐛 Bug Fixes
- scCODA and tascCODA results in
varmcan be written withwrite_h5muagain, sincesummary_preparereturns plain float64 DataFrames instead of theSummaryDataFramewith nullableFloat64columns that arviz 1.x returns (#1198) @Zethson DistanceTesttakeslayer_keyfor metrics that are not precomputed, such aseuclidean,mseandmean_absolute_error, which raised unlessobsm_keywas set, and accepts sparse layers there (#1201) @kkkklck