T cell co-inhibitory pathways¶
This notebook provides a reproducible analysis for a model of the T cell inhibitor checkpoints.
It contains the following results:
- identify stable states
- identify trap spaces
- propagation of CTLA4 and PD1 values through the model
Authors: Celine Hernandez, Aurélien Naldi
In [1]:
import ginsim
import biolqm
from colomoto_jupyter import tabulate
import pandas as pd
import numpy as np
In [2]:
# Load and view the model
gs_model = ginsim.load("Hernandez_TcellCheckPoints_13april2020.zginml")
lqm_model = ginsim.to_biolqm(gs_model)
ginsim.show(gs_model, save="figure_interaction_graph_model")
Helper functions¶
Here we define some helper functions and color mapping rules to perform value propagation and visualize the result.
In [3]:
# Transforms a dictionary into a dash-like pattern used for space restrictions.
# If a model has 4 components A, B, C, D in this order,
# {A:0, D:1} => "0--1"
def dash_pattern(model, dict_vals):
specific_comps = dict_vals.keys()
str_pattern = ""
for comp in model.getComponents():
if comp.toString() in specific_comps:
str_pattern += str(dict_vals.get(comp.toString()))
else :
str_pattern += "-"
return(str_pattern)
def restrict_model(model, **dict_vals):
pattern = dash_pattern(lqm_model, dict_vals)
return biolqm.restrict(lqm_model, pattern)
def fill_fixed(data, names, functions, mddman):
all_values = [f for f in functions]
for comp, func in zip(names, functions):
if mddman.isleaf(func): data[comp] = func
else: data[comp] = -1
def get_fixed_pattern(all_names, model, as_dict=False):
# Build a container for the results
pattern = {key: 100 for key in all_names}
# Model manager and core components
mddman = model.getMDDManager()
core_components = [node.getNodeID() for node in model.getComponents()]
extra_components = [node.getNodeID() for node in model.getExtraComponents()]
# 1/ Non-extra values: if the model was not reduced, core components may also contain fixed values
fill_fixed(pattern, core_components, model.getLogicalFunctions(), mddman)
# Special value for input components
for node in model.getComponents():
if node.isInput():
pattern[node.getNodeID()] = -2
# 2/ Extra values : only available after reduction/percolation
# Functions of each component
fill_fixed(pattern, extra_components, model.getExtraLogicalFunctions(), mddman)
if as_dict: return pattern
return pd.Series(pattern, dtype=np.byte).values.tobytes()
def compare_fixed_pattern(all_names, model1, model2, as_dict=False):
pattern1 = get_fixed_pattern(all_names, model1, as_dict=True)
pattern2 = get_fixed_pattern(all_names, model2, as_dict=True)
pattern = {}
for c in pattern1:
v1 = pattern1[c]
v2 = pattern2[c]
if v1 == v2: pattern[c] = v1
elif v1 < 0: pattern[c] = 10 + v2
elif v2 < 0: pattern[c] = 20 + v1
else: pattern[c] = 100
if as_dict: return pattern
return pd.Series(pattern, dtype=np.byte).values.tobytes()
def show_fixed_comparison(gs_model, restricted_model1, restricted_model2, styler, save=None):
name_components = [ n.getId() for n in gs_model.getNodeOrder() ]
pattern = compare_fixed_pattern(name_components, restricted_model1, restricted_model2)
styler.setState( pattern )
return ginsim.show(gs_model, style=styler, save=save)
def show_fixed(gs_model, restricted_model, styler, save=None):
name_components = [ n.getId() for n in gs_model.getNodeOrder() ]
fixed_pattern = get_fixed_pattern(name_components, restricted_model)
styler.setState(fixed_pattern)
return ginsim.show(gs_model, style=styler, save=save)
# Define color mapping rules
# Style for a single fixed pattern
styler_fixed = ginsim.lrg_style(gs_model)
styler_fixed.mapState2Color(0, 200, 25, 25)
styler_fixed.mapState2Color(1, 100, 175, 100)
styler_fixed.mapState2Color(2, 100, 225, 100)
styler_fixed.mapState2Color(-1, 255, 255, 255)
styler_fixed.mapState2Color(-2, 175, 175, 175)
# Style for comparing two patterns
styler_comp = ginsim.lrg_style(gs_model)
styler_comp.mapState2Color(-2, 175, 175, 175) # INPUT: gray
styler_comp.mapState2Color(0, 255, 255, 180) # OFF in both: light yellow
styler_comp.mapState2Color(1, 255, 180, 120) # ON in both: light orange
styler_comp.mapState2Color(2, 255, 180, 120) # HIGH in both: light orange
styler_comp.mapState2Color(-1, 255, 255, 255) # FREE in both: white
styler_comp.mapState2Color(10, 200, 255, 200) # OFF in the first: light green
styler_comp.mapState2Color(11, 125, 200, 125) # ON in the first: dark green
styler_comp.mapState2Color(20, 200, 200, 255) # OFF in the second: light green
styler_comp.mapState2Color(21, 125, 125, 200) # ON in the second: dark green
styler_comp.mapState2Color(100, 255, 180, 180) # Other (different values?): red
Identify stable states¶
In [4]:
# Model simplification:
# * Fix all inputs to their inactive state.
fixed = { str(n):0 for n in lqm_model.getComponents() if n.isInput()}
# Keep 'i_pMHCII_binding', 'i_pMHCII_agonist', (not 'i_pMHCII_dose'), 'i_pMHCII_affinity', 'i_CD80' as proper inputs
for n in ('i_pMHCII_binding', 'i_pMHCII_agonist', 'i_pMHCII_affinity', 'i_CD80', 'i_CD274'):
del(fixed[n])
# Fix inputs to their active state
fixed['CTLA4'] = 0
In [5]:
# Apply the simplifications defined above to reduce the model
m_reduced = biolqm.perturbation(lqm_model, " ".join([ "%s%%%s" % (k,v) for (k,v) in fixed.items() ]))
m_reduced = biolqm.reduce(m_reduced, ":fixed :purge :no-extra")
In [6]:
# Define reporter components
reporter_components = ["TCRalphabeta", "CD28", "NFAT_nuc", "FOS", "JUN", "NF_KB", "IL2", "Proliferation", "DGKA", "Anergy", "Quiescence"]
fps = biolqm.fixpoints(m_reduced, autoconvert=False
).project_on_components( reporter_components )
df = biolqm.states_to_dataframe(fps)
df
Out[6]:
| TCRalphabeta | CD28 | Quiescence | Proliferation | NF_KB | NFAT_nuc | DGKA | Anergy | FOS | JUN | IL2 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 1 | 0 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 2 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 3 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| 4 | 1 | 1 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 |
| 5 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 |
| 6 | 1 | 1 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 0 |
| 7 | 1 | 1 | 0 | 1 | 1 | 1 | 0 | 0 | 1 | 1 | 1 |
Identify trap spaces (skipped)¶
In [7]:
# Define reporter components
reporter_components = ["TCRalphabeta", "CD28", "NFAT_nuc", "FOS", "JUN", "NF_KB", "IL2", "Proliferation", "DGKA", "Anergy", "Quiescence"]
if False:
fps = biolqm.trapspaces(m_reduced, autoconvert=False
).project_on_components( reporter_components )
biolqm.states_to_dataframe(fps)
Percolate CTLA4 and PD1 values¶
No constrain¶
First we want to visualize fixed values on the model without any constraint. Fixed values will appear in white. This will be used as a reference to see if percolating a value will change anything from this default setting.
In [8]:
# Reduce fixed values
lqm_model_fixed_reduced = biolqm.reduce(lqm_model, ":fixed :purge")
show_fixed(gs_model, lqm_model_fixed_reduced, styler_fixed)
Out[8]:
CTAL4:0¶
In [9]:
# Percolate a value of 0 for CTLA4
lqm_model_restricted_ctla4_0 = restrict_model(lqm_model, CTLA4=0)
show_fixed(gs_model, lqm_model_restricted_ctla4_0, styler_fixed)
Out[9]: