NEWS

---

scTenifoldKnk 2.0.0

New features

• Direction of the response. knockoutDirection() predicts whether each gene goes up or down after a knockout, from the heat kernel of the WT gene-gene correlation matrix of log1p(CPM) expression. scTenifoldKnk() reports it by default (dr_direction = TRUE, dr_directionT = 5) as the direction and directionScore columns of diffRegulation, and as perturbationDirections in transcriptome-wide mode. dr_directionRegressLibSize = TRUE (off by default) regresses log library size out of each gene first, for data sets in which nearly all genes are predicted down.
• Heat manifold alignment. hkManifoldAlignment() computes the heat kernel of the WT network once (heatKernel()) and reads every knockout from it, so the cost of a transcriptome-wide screen no longer scales with the number of knockouts. Selected with ma_method = "heat" (ma_heatT = 10).
• perturbationMap() ranks the knockouts of a transcriptome-wide run by the similarity of their signed profiles to a reference signature, such as a disease-vs-control comparison of the same cell type, and draws them on a map.

Changes in behaviour

• diffRegulation has eight columns instead of six (direction, directionScore). Set dr_direction = FALSE for the 1.x table.
• Transcriptome-wide mode (transcriptomeWide = TRUE) uses the heat manifold alignment by default, so its distances differ from 1.x. Set ma_method = "manifold" to recover them. Single and multi-gene knockouts still use the manifold alignment by default and give the same distances, Z-scores and p-values as 1.1.5.

Documentation

• README rewritten: quality-control recommendations (removal of ribosomal and mitochondrial genes), use in tissues and disease data, and the limitations of the predicted direction.

scTenifoldKnk 1.1.5

• The knocked-out genes are left out of the expectation of the differential regulation test (#45).

scTenifoldKnk 1.1.4

• Distances at the level of floating-point noise are ignored by dRegulation().

scTenifoldKnk 1.1.3

• Warning when the networks of the requested genes do not fit in memory (#10).

scTenifoldKnk 1.1.2

• Multi-gene knockout: several genes in gKO are knocked out together (#47).
• plotKO() enrichment works without attaching enrichR (#43); data.frame input is rejected (#44).

scTenifoldKnk 1.1.1

• seed argument for reproducible results (#48); warning when the knocked-out genes have no outgoing edges.

scTenifoldKnk 1.1.0

• Transcriptome-wide perturbation mode and Efron's empirical null for the differential regulation test.