Compiler¶

In [41]:
from urllib.request import urlopen
import json
import numpy as np
from scipy.interpolate import interp1d

import matplotlib.pyplot as plt
%matplotlib inline

figsize=(6, 4.5)

Obtaining model and experimental data¶

Cello is a genetic circuit compiler. It compiles circuits written in a hardare description language (a sublanguage of Verilog) into NOR and NOT gates, to then finally generate DNA with the genetic circuit in it. Check out the paper by Nielsen et al. for a description of the compiler.

We will use modeling and experimental data from the Cello v2 compiler in the form of a UCF file. Let's load it into a variable so that we can work with it.

In [42]:
# fetch it form this url
url = "https://github.com/CIDARLAB/Cello-UCF/raw/cfb4de1cf125a6fb5af1971730844ea5d0e91623/files/v2/ucf/Eco/Eco1C1G1T1.UCF.json"
  
# get & parse
response = urlopen(url)
ucf_data = json.loads(response.read())

# if stored locally use this:
#
# with open('Eco1C1G1T1.UCF.json','r') as f:
#     s = f.read()
# ucf_data = json.loads(s)

And check out what is inside. It is a list that starts with a header that includes general information from the used bacteria to experimental conditions under which data was obtained:

In [43]:
print(ucf_data[0])

{'collection': 'header', 'description': 'TetR homologs: PhlF, SrpR, BM3R1, HlyIIR, BetI, AmtR, AmeR, QacR, IcaRA, LitR, PsrA, LmrA', 'version': 'Eco1C1G1T1', 'date': 'Thu Dec 24 00:56:10 EST 2015', 'author': ['Bryan Der', 'Timothy Jones'], 'organism': 'Escherichia coli NEB 10-beta', 'genome': 'NEB 10 ∆(ara-leu) 7697 araD139 fhuA ∆lacX74 galK16 galE15 e14- φ80dlacZ∆M15  recA1 relA1 endA1 nupG  rpsL (StrR) rph spoT1 ∆(mrr-hsdRMS-mcrBC)', 'media': 'M9 minimal media composed of M9 media salts (6.78 g/L Na2HPO4, 3 g/L KH2PO4, 1 g/L NH4Cl, 0.5 g/L NaCl, 0.34 g/L thiamine hydrochloride, 0.4% D-glucose, 0.2% Casamino acids, 2 mM MgSO4, and 0.1 mM CaCl2; kanamycin (50 ug/ml), spectinomycin (50 ug/ml)', 'temperature': '37', 'growth': 'Inoculation: Individual colonies into M9 media, 16 hours overnight in plate shaker.  Dilution: Next day, cells dilute ~200-fold into M9 media with antibiotics, growth for 3 hours.  Induction: Cells diluted ~650-fold into M9 media with antibiotics.  Growth: shaking incubator for 5 hours.  Arrest protein production: PBS and 2mg/ml kanamycin.  Measurement: flow cytometry, data processing for RPU normalization.'}

To get an overview on all the elements stored in the UCF file, let's see the collection names.

In [44]:
print([d['collection'] for d in ucf_data])

['header', 'measurement_std', 'logic_constraints', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'motif_library', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'gates', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'models', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'structures', 'functions', 'functions', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'parts', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'functions', 'device_rules', 'circuit_rules', 'genetic_locations']

A NOR gate from the library¶

The library contains several gates as can be seen in the collection variables. Le us study one such gate: the S3_SrpR.

In [56]:
gates_lib = [d for d in ucf_data if d['collection'] == 'gates']

gate_name = 'S3_SrpR'  # 'L1_LitR'
gate = [g for g in gates_lib if g['name'] == gate_name][0]

print(gate)

{'collection': 'gates', 'name': 'S3_SrpR', 'system': 'TetR', 'group': 'SrpR', 'regulator': 'SrpR', 'gate_type': 'NOR', 'color': '006838', 'model': 'S3_SrpR_model', 'structure': 'S3_SrpR_structure'}

In fact all gates are NOR gates as we can check:

In [57]:
print( [g['gate_type'] for g in gates_lib] )

['NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR', 'NOR']

Structure of the gate¶

It's structure, i.e., the genetic design, is obtained as follows:

In [58]:
structure_lib = [d for d in ucf_data if d['collection'] == 'structures']
structure = [s for s in structure_lib if s['name'] == gate['structure']][0]
print(structure)

{'collection': 'structures', 'name': 'S3_SrpR_structure', 'inputs': [{'name': 'in1', 'part_type': 'promoter'}, {'name': 'in2', 'part_type': 'promoter'}], 'outputs': ['pSrpR'], 'devices': [{'name': 'S3_SrpR', 'components': ['#in1', '#in2', 'S3_SrpR_cassette']}, {'name': 'S3_SrpR_cassette', 'components': ['RiboJ10', 'S3', 'SrpR', 'ECK120029600']}]}

Model of the gate¶

It's model is obtained by

In [59]:
model_lib = [d for d in ucf_data if d['collection'] == 'models']
model = [m for m in model_lib if m['name'] == gate['model']][0]
print(model)

{'collection': 'models', 'name': 'S3_SrpR_model', 'functions': {'response_function': 'Hill_response', 'input_composition': 'linear_input_composition', 'toxicity': 'S3_SrpR_toxicity', 'cytometry': 'S3_SrpR_cytometry'}, 'parameters': [{'name': 'ymax', 'value': 2.1, 'description': 'Maximal transcription'}, {'name': 'ymin', 'value': 0.004, 'description': 'Minimal transcription'}, {'name': 'K', 'value': 0.06, 'description': 'Dissociation equilibrium constant'}, {'name': 'n', 'value': 2.8, 'description': 'Cooperativety'}, {'name': 'alpha', 'value': 0.3337025658634261}, {'name': 'beta', 'value': 0.09255808434135396}]}

It says that a Hill response is used as a model with the additional specification that the inputs are composed linearly. We will check this out in a second as follows:

In [60]:
functions_lib = [d for d in ucf_data if d['collection'] == 'functions']

print('The first two elments describe these theoretical models:\n')
print(functions_lib[0])
print()
print(functions_lib[1])

The first two elments describe these theoretical models:

{'collection': 'functions', 'name': 'Hill_response', 'equation': 'ymin + (ymax - ymin) / (1.0 + (x / K)^n)', 'variables': [{'name': 'x', 'map': '#//model/functions/input_composition'}], 'parameters': [{'name': 'ymax', 'map': '#//model/parameters/ymax'}, {'name': 'ymin', 'map': '#//model/parameters/ymin'}, {'name': 'K', 'map': '#//model/parameters/K'}, {'name': 'n', 'map': '#//model/parameters/n'}]}

{'collection': 'functions', 'name': 'linear_input_composition', 'equation': 'x2 + x1', 'variables': [{'name': 'x1', 'map': '#//structure/inputs/in1/model/functions/response_function'}, {'name': 'x2', 'map': '#//structure/inputs/in2/model/functions/response_function'}]}

Measurements¶

Intrestingly there are even measurments that give much greater detail. Remember that the steady state input-output behavior is deterministic with Hill functions. In fact, measurments show that there is significant noise on output behavior.

To see this, let's look at measurements: The measurements are (normalized) histograms of steady-state input-output measurements in terms of Relative Promotor Units (RPUs). Details are described in the supplementary material of Nielsen et al.. For example, Figure S30 shows how distributions are probagated and Section VII.C.9. “gate_cytometry” states that RPUs in UCFs are measured on the same RPU scale to allow for propagation between gates.

Let's plot the response distribution for 3 inputs to such a component.

In [61]:
component = functions_lib[40]  # change to see other components: e.g. 30 for 'L1_LitR'

name = component['name']
measurements = component['table']

plt.figure(figsize=figsize)
plt.title(name)
plt.xlabel('RPU (consistent over all distributions)')

# Show some distributions
d = measurements[0]
plt.plot(d['bin'], d['output'], 'ro-', label=d['x'])

middle = int( len(measurements) / 2 )
d = measurements[middle]
plt.plot(d['bin'], d['output'], 'go-', label=d['x'])

d = measurements[-1]
plt.plot(d['bin'], d['output'], 'bo-', label=d['x'])

plt.xlim([0,50])
plt.legend()

# Let's check if the last one was really a distribution
print('Checking if distribution: this should be 1')
print(sum(d['output']))

Checking if distribution: this should be 1
1.000000000000001
No description has been provided for this image

The simplified input-output behavior¶

Using what we learned in Lecture 3, we plot the steady state input-output function of the NOR gate.

In [62]:
# Regulation functions
def nor(x, y):
    inputs = x + y  # linear input combination
    params = {}
    for p in model['parameters']:
        params[p['name']] = np.float64(p['value'])
        print(f"{p['name']}: {p['value']}")
    
    # using linear input combination
    Hill_equation = params['ymin'] + (params['ymax'] - params['ymin']) / (1.0 + (inputs / params['K'])**params['n'])                          
    
    # uncomment to use the one from Lecture 3
    # Hill_equation = params['ymin'] + (params['ymax'] - params['ymin']) / \
    #     ( (1.0 + (x / params['K'])**params['n']) * (1.0 + (y / params['K'])**params['n']) )                      

    return Hill_equation
In [63]:
# this is just copied from Lecture 3
def contour_fun(x, y, z, title=None):
    """
    Make a filled contour plot given x, y, z
    in 2D arrays.
    """
    plt.figure()
    CS = plt.contourf(x, y, z)
    cbar = plt.colorbar(CS)
    if title is not None:
        plt.title(title)

# Get x and y values for plotting
x = np.linspace(0, 0.1, 200)
y = np.linspace(0, 0.1, 200)
xx, yy = np.meshgrid(x, y)

# plot it
contour_fun(xx, yy,
            nor(xx, yy),
            title=f"{gate_name} gate")

ymax: 2.1
ymin: 0.004
K: 0.06
n: 2.8
alpha: 0.3337025658634261
beta: 0.09255808434135396
No description has been provided for this image

Model based on experimental data¶

To simplify this undertaking, we only model the expected input-output behavior.

In [65]:
# compute the mean
x = []
y = []
for d in measurements:
    x += [ d['x'] ]
    y += [ np.average(d['bin'], weights=d['output']) ]
    
plt.figure(figsize=figsize)
plt.loglog(x,y,'o-')

# let's create a function from it that we can use
# we go for linear interpolation
# and enable extrapolation to be able to use it for values
# outside of the given data range
Hill_measured = interp1d(x,y, fill_value='extrapolate')

# and show it
plt.figure(figsize=figsize)
x_new = np.linspace(0,2,100)
plt.loglog(x_new, Hill_measured(x_new), 'g-')
plt.loglog(x, y, 'o')
Out[65]:
[<matplotlib.lines.Line2D at 0x7f9e2bf2cac0>]
No description has been provided for this image
No description has been provided for this image

We are now ready to use it as a NOR model. Mind that we do not have experimental data on the individual inputs, so we go for the linear combination as proposed in the Cello v2 model of the NOR.

In [54]:
def nor_experimental(x, y):
    inputs = x + y  # linear input combination
    
    # using linear input combination
    Hill_equation = Hill_measured(inputs)                                              

    return Hill_equation

# Get x and y values for plotting
x = np.linspace(0, 0.1, 200)
y = np.linspace(0, 0.1, 200)
xx, yy = np.meshgrid(x, y)

# plot it
contour_fun(xx, yy,
            nor_experimental(xx, yy),
            title=f"{gate_name} gate (experimental) with linear input combination")
contour_fun(xx, yy,
            nor(xx, yy),
            title=f"{gate_name} gate (Hill model) with linear input combination")

ymax: 4.3
ymin: 0.07
K: 0.05
n: 1.7
alpha: 0.3748404992150903
beta: 0.43564010137683185
No description has been provided for this image
No description has been provided for this image

Tasks¶

  1. Plot all deterministic Hill functions for all NOR gates in log-log plots and discuss what you see. You can plot them for a single input value and not two.

  2. Propagate determinsitic input-output values from inputs to an output. Write a function that allows us to call progagate(circuit,i) where circuit is a linear circuit of NOR gates but with a single input (INV gates). The circuit may, e.g., be a list of str with NOR names. Argument i is an input value in RPUs. The output of the function is the output of the circuit in RPUs.

  3. Do the same (1. and 2.), but with distributions as the output based on the measurements in the UCFs. For 1', you can take a look at seaborn.violinplot.

  4. Write a compiler that, given a length of the INV chain, returns a "good" circuit (in the above sense). Explain what you chose as a criterion for "good" and why.

  5. Write a compiler that can synthesize Boolean circuits. What are its limits? What quality measures is it optimized for?

In [ ]: