Changelog • scDiagnostics

CHANGES IN VERSION 1.0.0

Add functionality to process SCE objects for PCA computation via the new processPCA() function.
Add functionality to downsample SCE objects for diagnostic functions.
Add new diagnostic functions (calculateTopLoadingGeneShifts(), compareMarkers() and calculateMMDPValue()).
Add graph integration diagnostic function in replacement of nearest neighbor diagnostic, calculateGraphIntegration().
Improve regressPC() function and plot method, which can now also regress against cell types and batches.
Improve normalization for plotMarkerExpression() diagnostic function.
Improve user control and options for plot methods.
Update vignettes to reflect all new changes.

Renamed gene shift function for consistency (previously calculateTopLoadingGeneShifts())
Added gene specification parameter to calculateGeneShifts()
Improved calculateGeneShifts() function, plot method, and color scheme

Added calculateReconstructionError() to detect out-of-distribution anomalies using cell-type-specific PCA reconstruction errors.
Replaced the 4 general-purpose vignettes with 4 vignettes built around the package’s manuscript: an introduction with a simulated ground-truth example, a benchmark of detectAnomaly()/calculateReconstructionError() against known ground truth on the Zeisel brain dataset, and two case studies (COVID-19 PBMC scRNA-seq and MERFISH spatial colitis data) walking through the project/detect/characterize diagnostic workflow.
Added new example datasets (zeisel_reference_data, zeisel_query_data, zeisel_benchmark_results, covid_reference_data, covid_query_data, merfish_reference_data, merfish_query_data) supporting the new vignettes.
Added plot.calculateReconstructionErrorObject() featuring robust visualization options (violin, boxplot, ridge, and ComplexHeatmap).
Upgraded detectAnomaly() to resolve the curse of dimensionality by allowing Isolation Forests to run on the union of query and reference Highly Variable Genes (via n_hvgs) when pc_subset = NULL.
Improved anomaly detection by switching default thresholding to a dynamic, data-driven Median Absolute Deviation method (threshold_method = "MAD", mad_multiplier = 2) across relevant functions.