Visualization

sfa.vis provides a small set of helpers for turning simulation outputs into visual representations of signal flow.

Computing graphics from a simulation

sfa.vis.compute_graphics annotates a NetworkX DiGraph with per-node fill colors and per-edge widths/signs/colors derived from an activity change and a signal-flow matrix. The result is a graph you can render with any NetworkX-aware drawing layer.

import sfa
from sfa.vis import compute_graphics
from sfa.analysis import analyze_perturb

# Run two conditions to get an activity-change vector `act` and
# signal-flow matrix `F`.
act, F = analyze_perturb(alg, data, targets=['MEK'])

dg = compute_graphics(
    F=F,
    act=act,
    A=data.A,
    n2i=data.n2i,
    lw_min=1.0,
    lw_max=10.0,
    pct_link=90,   # Percentile used to scale link widths.
    pct_act=50,    # Percentile used to scale node fills.
)

dg is a NetworkX.DiGraph. Each node has FILL_COLOR, BORDER_WIDTH, and BORDER_COLOR attributes; each edge has SIGN, FILL_COLOR, and WIDTH attributes. The fill color blends from white toward red for up-regulated nodes and toward blue for down-regulated nodes, scaled by the percentile of activity change you pass in. Edge widths are mapped from \(\log_{10}|F|\) clipped at the pct_link percentile.

The pre-existing nodes of a graph object can be reused by passing a dg= argument; otherwise a fresh DiGraph is created and populated from n2i.

Integration with SFV

For interactive rendering, SFA can hand its annotated graph off to SFV (Seamless Flow Visualization). sfa.vis.sfv.sfv.visualize_signal_flow does the same coloring/widening work as compute_graphics, but writes the styles directly into an sfv.graphics.Network object using Qt-based color objects and header shapes.

from sfa.vis.sfv.sfv import visualize_signal_flow

visualize_signal_flow(
    net,            # sfv.graphics.Network
    F=F, act=act, A=data.A, n2i=data.n2i,
    color_up=None,  # Defaults to red.
    color_dn=None,  # Defaults to blue.
    show_label=True,
    show_act=True,
    pct_act=50,
    pct_link=90,
)

This subpackage requires qtpy and sfv to be available; both are optional dependencies of sfa itself.

Building a Data object from an SFV network

If you start from an SFV network instead of a SIF file, the helper sfa.vis.sfv.sfv.create_from_graphics constructs a one-off sfa.Data subclass whose topology is taken from the graphics object.

from sfa.vis.sfv.sfv import create_from_graphics

data = create_from_graphics(net, abbr="MyNet", inputs={'EGF': 1.0})

The returned object only carries network topology; experimental dataframes (df_conds, df_exp, df_ptb) are left as None, so it is ready to use with sfa.AlgorithmSet-created algorithms but not with random batch simulators that compare against experimental data.

Plot helpers

sfa.plot collects a few matplotlib-based plots that are useful for inspecting simulation results:

Symbol	Description
sfa.plot.Heatmap	Heatmap of an n × m activity or accuracy matrix.
sfa.plot.BatchResultTable	Tabular view of Algorithm.result.df_sim.
sfa.plot.ConditionTable	Tabular view of Data.df_conds.
sfa.plot.HierarchicalClusteringTable	Hierarchically clustered table view.
sfa.plot.siplot	SPLO–Influence bar grid (see Control).