Shared nearest neighbors via fast Jaccard index calculation

Computes a shared nearest neighbor (SNN) graph by calculating Jaccard similarity between cells based on their k-nearest neighbors. The Jaccard index J(A,B) = |A∩B| / |A∪B| measures the overlap of neighbor sets. This implementation uses sparse matrix operations for efficiency.

Usage

fast.jaccard.r(.adj.matrix, .prune = 1/15)

Arguments

.adj.matrix

A non-symmetric sparse adjacency matrix from get.adj.matrix, where entry (j,i)=1 indicates j is among i's nearest neighbors (neighbors in rows, cells in columns).

.prune

A numeric threshold (default 1/15 ≈ 0.067) for pruning weak edges. Jaccard values below this are set to zero for sparsity.

Value

A symmetric sparse matrix of Jaccard indices representing the SNN graph. Higher values indicate greater neighbor overlap. Entry (i,j) gives the Jaccard similarity between cell i and cell j based on their shared k-nearest neighbors. Matrix is pruned to remove weak connections and forced symmetric.