`genomeblocks.architecture`

Architecture(graph_tool.Graph)

`Architecture(graph_tool.Graph)`

Undirected by default (some builders accept directed=True). Every vertex has vp.uid; every edge has ep.w (weight), ep.n (normalized), ep.d (distance).

Constructor

Architecture(name: str | None = None)

Properties & shortcuts

Name	Description
`n_loci`	`num_vertices()`.
`n_links`	`num_edges()`.
`index`	`{uid → Vertex}`.

Lookup

arch[uid]                    # {neighbor_uid: ep.w[edge]}
arch[(uid1, uid2)]           # {ep_name: value} for the connecting edge
uid in arch                  # bool
len(arch)                    # vertex count

Construction

Architecture.make(loci, bedpe, *, name="Skeleton",
                  r=2500, dmax=1e9, verbose=True) -> Architecture
# Build from BEDPE loops (parsed via bedpe.read_bedpe); r = ±radius around
# anchor midpoints for CRE mapping, dmax drops far-apart loops.

Weight assignment & normalization

arch.add_mcool(loci, mcool, *, resolution=None,
               name="w", verbose=True) -> Architecture
# Assign Hi-C pixel sums; distributes across overlapping edges per bin pair.

arch.normalize(loci, *, source="w", name="n",
               verbose=True) -> Architecture
# Fit w ≈ C·d^-α power law; write O/E to ep.n (or `name`), distances to ep.d.

arch.prune(*, dist_prop="d", verbose=True) -> Architecture
# Drop zero-distance (co-located) edges. Call once, after normalize.

Annotation

arch.annotate(loci, genes, *, key="n", name="gene", verbose=True) -> Architecture
# vp.annot = region class; vp[name] = nearest gene (promoters) or the gene of the
# top-`key`-weight promoter neighbour (other CREs). Run normalize first.

Hub genes

arch.strength(key="n", name="strength", *, verbose=True) -> Architecture
# Node strength: vp[name] = sum of incident ep[key] (no normalization).

arch.elbow(key, *, verbose=True) -> (cutoff, sorted_uids)
# Slope-1 knee over vp[key] descending; cutoff = number of hubs.

arch.prime_hubs(key="n", gene="gene", *, verbose=True) -> dict
# dict keys: 'prime_genes', 'promoter_genes', 'enhancer_genes',
#            'hub_uids', 'cutoff', 'promoter_uids', 'enhancer_uids'

Subsetting & set operations

arch.copy() -> Architecture                           # deep-copy all props
arch.subgraph(filter_func=None, vp_name=None, vp_values=None,
              uids=None, name=None) -> Architecture   # one-of filter modes

arch | other                                           # vertex+edge union
arch & other                                           # common vertices + common edges

Visualization

Drawing lives in genomeblocks.architecture_draw (separate module, matplotlib):

from genomeblocks.architecture_draw import draw

draw(arch, loci, region, *,
     merge_distance=None,     # collapse nearby loci into single nodes
     vertex_size_by=None,
     edge_width_by='w',
     vertex_color=None,       # str hex | vp_name | PropertyMap | None
     edge_color='#CCCCCC',
     figsize=(12, 8),
     layout='spring',         # 'spring' | 'circular' | 'kamada_kawai'
     show_labels=True,
     label_prop='gene',
     ax=None,
     **kwargs) -> matplotlib axis

Serialization

arch.to_frame() -> pandas.DataFrame    # one row per vertex, one col per vp
pickle.dumps(arch)                      # full property round-trip via __getstate__/__setstate__

genomeblocks.architecture

Table of contents

Architecture(graph_tool.Graph)