automatic compilation from high-level bio-languages to ... › talks › iwbda-2010.pdf ·...

Automatic Compilation from High-Level Bio-Languages to Genetic Regulatory Networks

Jacob Beal, Ting Lu, Ron Weiss

IWBDA, June 2010

Goal: High-Level Biological Design

(def band-detector (signal lo hi) (and (> signal lo) (< signal hi)))

(let ((v (diffuse (aTc) 0.8 0.05))) (green (band-detect v 0.2 1)))

High-LevelBio-focusedLanguage

OptimizeCompile

Genetic RegulatoryNetwork

Sim

ula te

Assem

ble

Tool Chain Vision

Outline

● Compositional Design● Motif-Based Compilation● Simulation Results

ribosome

Computation via Transcription Network

DNA

RNARNA polymerase

promoter

regulatoryprotein

ProteinDecay

ribosome

Computation via Transcription Network

Stablizes at decay = production

DNA

RNA

promoter

regulatoryprotein

ProteinDecay

Signal = Concentration

RNA polymerase

Alternatives:

PoPS

RNA concentration

X

Abstract GRN Design Space

Y

Z

XInputs:[none]

,,

X

Abstract GRN Design Space

Y

Z

X

d [ X ]=Ri⋅1K Y−1[Y ]/DY

H Y

1[Y ]/DY H Y ⋅ 1K Z [Z ]/DZ

H Z

1[Z ]/DZ H Z −log2/ t X [ X ]

K ∈ [2,1000]

D ∈ [10,1000] nM

[X] ∈ [0,1000] nM

t > 300 s

H ∈ [1,4]

R < 10 nM/s

Inputs:[none]

,,

KY D

Y H

Y

KZ D

Z H

Z

tX

Ri

Major Challenge: Interference

● Effective part characterics changed by:● Cellular context (endogenous pathways, synthetic parts)● Expression noise

● Our approach: noise-rejection

● Digital - static discipline: Vlow,out

< Vlow,in

< Vhigh,in

< Vhigh,out

But part variance makes a uniform standard impossible!

Vlow,in

Vhigh,in

Vlow,out

Vhigh,out

[In]

[Out]

Parameterized Standards Families

● Identify “standards family” parameter relation

● Create library of characterized part variants

● Adjust part choice to match on junctions

Experimental Input to Family Relation

[Basu & Weiss, '05]

Model constrained by characterization experiments...

[Karig, '07][Subramanian, '08]

[Weiss, '01]

Draft Simulation-Based Standards Family Generation

Need: K,H,D,R,t,max [X]

● Choose max [X]

● Assuming large K, amplification of 0.5K when [X]/D ≈ 1

● H → D, static discipline (higher H is better)

● Steady state max: [X]=production*t/log(2)

● production ≈ K*RExample:

max [X] = 500 nM H = 3R = 0.193 t = 1800 sK = 101 D = 190

Vlow,out

Vhigh,in

Vhigh,out

Vlow,in



● Choose max [X]





max [X] = 500 nM H = 3R = 0.193 t = 1000 sK = 101 D = 80

Vlow,out V

high,in

Vhigh,out

Vlow,in



● Choose max [X]





max [X] = 500 nM H = 3R = 0.193 t = 1000 sK = 101 D = 80

Driver for experimental part creation & characterization!

Vlow,out V

high,in

Vhigh,out

Vlow,in

Outline


Motif-Based Compilation

● High-level primitives map to GRN design motifs● e.g. logical operators:

(primitive not (boolean) boolean :bb-template ((P 0.193 R- arg0 RBS outputs T)))

arg0 output


● High-level primitives map to GRN design motifs● e.g. logical operators, actuators:

(primitive green (scalar) scalar :side-effect :bb-template ((P R+ arg0 RBS GFP outputs T)))

GFP outputarg0


● High-level primitives map to GRN design motifs● e.g. logical operators, actuators, sensors:

(primitive IPTG () scalar :bb-template ((P 0.193 RBS LacI T) (rxn LacI (IPTG 180000) -> LacI*)

(P 0.193 R- LacI RBS outputs T)))

outputLacI

IPTG


● Functional program gives dataflow computation:

● Operators translated to motifs:

● Standards family sets chemical constants● Optimizers simplify network

(green (not (IPTG)))

LacI A

IPTG

B GFP

IPTG not green




IPTG not green


● Operators translated to motifs:● Standards family sets chemical constants

IPTG not green



not greenLacI A

IPTG



greenLacI A

IPTG

B



LacI A

IPTG

B GFP C


● Optimizers simplify network

LacI A

IPTG

B GFP C



LacI A

IPTG

B GFP



LacI A

IPTG

GFP

Outline


Simulation Results

● Prototype compiler generates GRNs that simulate correctly for a limited language subset

● Example: 2-bit adder

a1a0

b1b0

x1x0

carry

Simulation Results



a1a0

b1b0

x1x0

carry

1

true

3

2

Simulation Results



a1a0

b1b0

x1x0

carry2

3

1

true

1

00

1

1

1

1

Simulation Results



a1a0

b1b0

x1x0

carryaTcIPTG

C4HSL

3OC12

HSL

blue

redgreen

2

3

1

true

1

00

1

1

1

1

Simulation Results


● Example: 2-bit adder(macro xor (a b) (muxor (muxand ,a (not ,b))

(muxand ,b (not ,a))))

(macro 2bit-adder (a1 a0 b1 b0) (all (green (xor ,a0 ,b0)) ; x_0 low bit (let ((c0 (muxand ,a0 ,b0))

(x1 (xor ,a1 ,b1))) (red (xor x1 c0)) ; x_1 high bit (blue (muxor (muxand x1 c0) ; carry bit (muxand ,a1 ,b1))))))

(2bit-adder (aTc) (IPTG) (C4HSL) (3OC12HSL))

Simulation Results

● Compiled 2-bit adder (unoptimized)● 60 signal chemicals● 52 regulatory regions

● Generated ODE simulation in MATLAB

00+00

Bit 0Bit 1Carry

10+01 11+10 01+01

On to optimization...

● Adapted classical techniques can be powerful:

Band detector optimization from [Beal & Bachrach, '08]

(def band-detector (signal lo hi) (and (> signal lo) (< signal hi))) (let ((v (diffuse (aTc) 0.8 0.05))) (green (band-detect v 0.2 1)))

Tool Chain Vision

Contributions

● Parameterized standards identify all chemical parameters that can produce digital logic in transcriptional networks.

● Prototype motif-based compiler automatically maps high-level programs into GRNs.

● Automatically generated MATLAB ODE simulations verify that GRNs implement program specification.

automatic compilation from high-level bio-languages to ... › talks › iwbda-2010.pdf ·...

Documents