nuget
api documents
Precomputed database for FELLA analysis. Contains the KEGG graph and all precomputed matrices needed for the three enrichment methods: hypergeometric, diffusion, PageRank.
This is equivalent to the FELLA.DATA S4 class in the R package. Building this database is a one-time operation per organism; it can be serialized and reloaded for subsequent analyses.
01 Syntax
02 Methods
| Name | Overloads | Summary |
|---|---|---|
| BuildHypergeomMatrix | 1 | Build the hypergeometric matrix (pathway-compound incidence). |
| FindReachableCompounds | 1 | Find all compound nodes reachable from a given node through the graph. |
| BuildDiffusionKernel | 1 | Build the diffusion kernel matrix. |
| BuildPagerankMatrix | 1 | Build the PageRank matrix. |
| BuildDiffusionNullStats | 1 | Precompute null distribution statistics for diffusion using normality approximation. |
| BuildPagerankNullStats | 1 | Precompute null distribution statistics for PageRank using normality approximation. |
| BuildAll | 1 | Build all precomputed data structures. |
03 Properties
| Name | Overloads | Summary |
|---|---|---|
| Graph | 1 | The KEGG multi-layer graph |
| BackgroundCompounds | 1 | Background compound IDs (all compounds in the graph) |
| HypergeomMatrix | 1 | Pathway-compound incidence matrix for hypergeometric test. |
| DiffusionKernel | 1 | Precomputed inverse of the regularized Laplacian for diffusion. |
| PagerankMatrix | 1 | Precomputed matrix for PageRank. |
| DampingFactor | 1 | Damping factor for PageRank (default 0.85) |
| Gamma | 1 | Regularization parameter gamma for diffusion (default 1.0) |
| DiffusionNullStats | 1 | Precomputed null distribution statistics for diffusion. |
| PagerankNullStats | 1 | Precomputed null distribution statistics for PageRank. |
04 Members
Build the hypergeometric matrix (pathway-compound incidence). For each pathway, identifies which compounds are reachable through the graph hierarchy (pathway -> module -> enzyme -> reaction -> compound).
Int32)Find all compound nodes reachable from a given node through the graph. Uses BFS traversal following edges.
Build the diffusion kernel matrix. K = (L + gamma * B)^{-1} where L is the unnormalized Laplacian, B is the diagonal "leak" matrix (1 for pathway nodes, 0 otherwise), and gamma is the regularization parameter.
The diffusion score for input vector y is: f = K * y
Build the PageRank matrix. PR_matrix = (1-d) (I - dM)^{-1} where M is the column-stochastic transition matrix and d is the damping factor.
PageRank scores for personalization vector p: pr = PR_matrix * p
Int32)Precompute null distribution statistics for diffusion using normality approximation. For each node i, computes E[f_i] and Var(f_i) under the null hypothesis that the input compounds are randomly sampled from the background.
Uses the analytical formulas: E[f_i] = (n_input / n_background) sum_j K[i,j] (for compound nodes j) Var[f_i] = (n_input / n_background) (1 - n_input/n_background) * sum_j K[i,j]^2 + correction terms
Int32)Precompute null distribution statistics for PageRank using normality approximation. Similar to diffusion but uses the PageRank matrix instead.
Int32, Boolean, Boolean, Boolean)Build all precomputed data structures.
The KEGG multi-layer graph
Background compound IDs (all compounds in the graph)
Pathway-compound incidence matrix for hypergeometric test. Rows = pathways, Columns = compounds. H[p,c] = 1 if compound c belongs to pathway p.
Precomputed inverse of the regularized Laplacian for diffusion. K = (L + gammaB)^{-1} Diffusion scores = K input_vector
Precomputed matrix for PageRank. PR = (1-d) (I - dM)^{-1} * p where M is the transition matrix, d is damping factor, p is personalization.
Damping factor for PageRank (default 0.85)
Regularization parameter gamma for diffusion (default 1.0)
Precomputed null distribution statistics for diffusion. For each node, stores (mean, variance) under the null distribution where input compounds are randomly sampled from the background.
Precomputed null distribution statistics for PageRank.