Infer Receptor Activity from NeighbourNet TF–Target Networks

This function infers receptor, transcription factor (TF), and target activity by combining a receptor–TF prior model (personalised PageRank) with NeighbourNet TF–target co-expression networks. Activity can be evaluated at the per-cell level or on meta-networks, and can be optionally filtered by receptor expression and prior gene regulatory network (GRN) evidence.

Usage

receptor.activity(
  seurat.obj,
  i = NULL,
  meta.network = FALSE,
  cutoff = NULL,
  check.receptor.expression = TRUE,
  check.gr.evidence = TRUE,
  t = 2,
  p = NULL,
  scale.ppr = TRUE,
  scale.network = TRUE,
  as.tfs = c("predictors", "responses"),
  receptors = NULL,
  tfs = NULL,
  targets = NULL,
  receptor.activity = c("cprod", "dist")
)

Arguments

seurat.obj: A Seurat object with a NNet.mod list stored in the misc slot. This list is created by run.nn.reg. To optionally infer receptor activity on meta-networks (set meta.network = TRUE), the NNet.mod should be updated by a subsequent run of build.meta.network.
i: Optional index of cells or meta-network components for which to compute receptor activity. If NULL, uses NNet.mod$cells for per-cell networks or all meta-network components when meta.network = TRUE.
meta.network: A logical indicating whether to compute activity on NeighbourNet meta-networks instead of per-cell networks. If TRUE, the "meta.network" assay is used and each component is treated as a meta-cell. Default is FALSE.
cutoff: A numeric value specifying the p-value threshold to prune insignificant co-expression values. If NULL, the default value is taken from NNet.mod$defaults.
check.receptor.expression: A logical indicating whether receptor activity should be down-weighted by receptor expression. Default is TRUE.
check.gr.evidence: A logical indicating whether TF–target edges should be filtered by prior GRN evidence from get.gr.adj. Default is TRUE.
t: A numeric value passed to get.gr.adj that controls the depth or order of TF–target adjacency used as prior GRN evidence. Default is 2.
p: A numeric value passed to get.prior.model specifying the quantile threshold for pruning the PPR matrix. If NULL, the default threshold stored in receptor.ppr$ltf is used. Default is NULL.
scale.ppr: A logical indicating whether to normalise the receptor–TF prior model (PageRank matrix) by receptor-wise norms. Default is TRUE.
scale.network: A logical indicating whether to normalise the TF–target co-expression network by target-wise norms before computing activity. Default is TRUE.
as.tfs: A character string indicating whether TFs are encoded as "predictors" (columns) or "responses" (rows) in the NeighbourNet networks. Default is "predictors".
receptors: A character vector of receptor genes to include. If NULL, all receptors present in the prior model returned by get.prior.model are used.
tfs: A character vector of TFs to include. If NULL, TFs are taken from the default gene set stored in NNet.mod$defaults.
targets: A character vector of target genes to include. If NULL, targets are taken from the default gene set stored in NNet.mod$defaults.
receptor.activity: A character string specifying how receptor–TF and TF–target information should be combined. Options are "cprod" (cross-product aggregration that measures cosine similarity, default) and "dist" (max-over-path aggregation).

Value

If multiple cells or meta-network components are requested (length(i) > 1 or i = NULL), the function returns a list with:

receptor.act: A matrix of receptor activity scores (receptors x cells/components).
tf.act: A matrix of TF activity scores (TFs x cells/components).
target.act: A matrix of target activity scores (targets x cells/components).

If a single cell or component is requested, the function returns a list with:

act.mat: A receptor–target activity matrix (receptors x targets).
ppr: The receptor–TF prior used after any scaling and expression weighting.
network: The TF–target network used for activity computation.

Details

This function integrates three hierarchical components to infer receptor activity: (i) a receptor–TF prior model obtained from get.prior.model, (ii) TF–target co-expression networks derived from NeighbourNet, and (iii) optional TF–target adjacency matrices from get.gr.adj as prior gene regulatory network (GRN) evidence. Together, these components define a receptor–TF–target signalling cascade that quantifies how upstream receptors potentially influence downstream target genes through TFs.

For each cell or meta-network component, the receptor–TF prior model encodes the probability of regulatory transmission from receptors to TF, while the TF–target network represents local co-expression relationships inferred by NeighbourNet. Receptor activity is computed by aggregating these two layers to estimate receptor influence on downstream targets, mediated by TFs. The resulting matrices describe receptor-, TF-, and target-level activity for each cell or component, providing an interpretable summary of upstream signalling dynamics.

Activity can be inferred from per-cell networks (default) or from meta-networks when meta.network = TRUE. The arguments scale.ppr and scale.network apply L2-normalisation to the prior model and co-expression network respectively, reducing the influence of highly connected nodes and ensuring comparable contribution across genes. The options check.receptor.expression and check.gr.evidence introduce two filters on receptor activity : the first enforces that receptors must be transcriptionally detected, while the second prunes TF–target co-expression to those supported by prior GRN knowledge.

The argument receptor.activity controls how the two layers are combined: "cprod" performs a cross-product aggregation equivalent to measuring cosine similarity between receptor–TF and TF–target profiles, while "dist" applies a max-over-path operation that highlights the strongest possible receptor–target link through any intermediate TF.

Examples