1 of 124

Flux Exponent Control

Predicts Metabolic Dynamics

from Network Structure

Fangzhou Xiao, Jingshuang (Lisa) Li, John C Doyle

UCSD, Caltech

American Control Conference

202305

2 of 124

Bioengineering: engineer cells as biomachines.

Biocontrol: How to formulate cells as controllable machines?

Essential foundation for a mature bioindustry.

Analyze and design cells like cars and power grids.

David Goodsell

Capture uniquely biological properties, not borrowing existing formulations from other disciplines.

3 of 124

Biocontrol: formulate cells as a controllable machine.

Gas laws

Transistors, band gap…

Electronics, radiation, amplifier...

Ohm’s law, …

Systems theory

Structures of interaction

Mechanical components

Mechanical machine

Newtonian,

Mass & Force

Steam engine

Thermo-

dynamics

Computer

Turing machine

Electrical circuits

Linear i/o systems

Communication network

Information channels

Components

Machines

Components

Machines

Lagrangian

4 of 124

Cells are metabolic machines.

Biocontrol: formulate cells as a controllable machine.

Molecules IN

Molecules OUT

Systems theory

Structures of interaction

Biomolecular reactions

Metabolic machines

Theory of

Flux exponent control (FEC)

Binding & Catalysis

Tool: Reaction order polyhedra (ROP)

David Goodsell

Components

Machines

Components

Machines

Specific goal: model metabolism dynamics.

5 of 124

Metabolism dynamics is only sparsely known ⇒ hard to model.

Lack knowledge about enzyme regulation.

substrate

(total)

product

(total)

You see this:

wikipedia

6 of 124

Metabolism dynamics is only sparsely known ⇒ hard to model.

You see this:

Lack knowledge about enzyme regulation.

More generally:

x, metabolite conc., variable

S, stoichiometry, known.

v, flux, unknown.

substrate

(total)

product

(total)

wikipedia

7 of 124

Flux control: cells are metabolic machines that control fluxes.

Stoichiometry, Known

Flux,

Unknown

8 of 124

Flux control: cells are metabolic machines that control fluxes.

Machine architecture:

Stoichiometry.

Machine control actions:

Flux control.

9 of 124

Flux control: cells are metabolic machines that control fluxes.

Lack knowledge about enzyme regulation.

You see this:

Machine architecture:

Stoichiometry.

Machine control actions:

Flux control.

More generally:

x, metabolite conc., variable

S, stoichiometry, known.

v, flux, unknown.

10 of 124

Constraint-based methods to model metabolism fill in unknown via optimization.

Specify detailed mechanisms.

Completely known.

Choose

biologically feasible fluxes.

11 of 124

Constraint-based methods to model metabolism fill in unknown via optimization.

e.g. Growth

Biological fluxes

Specify detailed mechanisms.

Completely known.

Choose

biologically feasible fluxes.

Flux control has too few constraints for dynamic fluxes.

12 of 124

Flux control has few constraints for dynamic fluxes.

Choose

stoichiometry-compatible fluxes.

Biological fluxes

Choose

biologically feasible fluxes.

13 of 124

Flux control has few constraints for dynamic fluxes.

Choose

stoichiometry-compatible fluxes.

e.g. Growth

Biological fluxes

Choose

biologically feasible fluxes.

Flux control (FC)

14 of 124

In practice, only steady state fluxes has enough constraints.

Choose

stoichiometry-compatible fluxes.

e.g. Growth

Biological fluxes

Flux control can only capture s.s. metabolism, because it ignores intrinsic dynamics.

Flux control (FC), incl. FBA

Flux control for steady state fluxes is flux balance analysis (FBA).

Choose static fluxes.

15 of 124

Dynamics in life, e.g. glycolysis.

Autocatalysis (e.g. glycolysis) is a core part of life,

From energy generation to growth.

Positive feedback.

Intrinsically unstable.

Require active regulation.

Glycolysis generates ATP from glucose.

Consume ATP.

Produce ATP.

Chance, Schoener, Elsaesser,

PNAS, 1964

Dano, Sorensen, Hynne,

Nature, 1999

16 of 124

Dynamics in life, e.g. glycolysis.

Autocatalysis (e.g. glycolysis) is a core part of life,

From energy generation to growth.

Glycolysis generates ATP from glucose.

Flux control: Trivial plant, ignores intrinsic dynamics

Consume ATP.

Produce ATP.

Chance, Schoener, Elsaesser,

PNAS, 1964

Dano, Sorensen, Hynne,

Nature, 1999

17 of 124

Flux control ignores metabolic fluxes are catalyzed by enzymes.

Machine architecture

Metabolic machines that control fluxes.

Metabolic fluxes have intrinsic dynamics without regulation!

They are catalyzed by enzymes!

18 of 124

Fluxes are controlled via binding’s regulation of enzyme activity.

19 of 124

Fluxes are controlled via binding’s regulation of enzyme activity.

How to formulate this mathematically?

20 of 124

Binding regulates fluxes’ exponents.

Cells control fluxes by binding;

Cells control fluxes’ exponents!

(Flux exponent control, FEC)

+

=

21 of 124

Cells control flux exponents, not fluxes themselves.

Machine architecture

Metabolic machines that control fluxes.

Metabolic machines that control fluxes’ exponents.

Idea of exponent regulation dates back to Michael Savageau’s S systems in 1970s.

Metabolic fluxes have intrinsic dynamics without regulation!

They are catalyzed by enzymes!

22 of 124

How to formulate flux exponent control mathematically.

Flux exponent control (FEC)

Flux control (FC)

Reference flux magnitude

Unregulated reaction dynamics

Regulation of flux exponents by binding

23 of 124

How to formulate flux exponent control mathematically.

Reference flux magnitude

Passive reaction dynamics

Regulation of flux exponents by binding

24 of 124

Intrinsic dynamics of metabolism: Autocatalysis.

Autocatalysis is a core part of life,

From energy generation to growth.

Positive feedback.

Intrinsically unstable dynamics.

Require active regulation.

Glycolysis: cells eating sugar.

Consume ATP.

Produce ATP.

25 of 124

Intuition: autocatalysis is like stick balancing.

Autocatalysis is a core part of life,

From energy generation to growth.

Positive feedback.

Intrinsically unstable dynamics.

Require active regulation.

Consume ATP.

Produce ATP.

26 of 124

Direct control of all fluxes is like no inertia.

Autocatalysis is a core part of life,

From energy generation to growth.

Positive feedback.

Intrinsically unstable dynamics.

Require active regulation.

Machine architecture

Machine control actions

Metabolic fluxes have unmodifiable intrinsic dynamics!

How to capture this?

27 of 124

FEC as a constraint-based approach.

In vector form:

Regulates flux exponents (reaction order) for static controller:

FBA

FEC

28 of 124

Flux exponent control (FEC) can constrain dynamic fluxes.

Biological fluxes

Choose

dynamic control of fluxes’ exponents.

(Correspond to regulation in binding networks)

Choose

stoichiometry-compatible fluxes.

Flux control (FC), incl. FBA

29 of 124

Flux exponent control (FEC) can constrain dynamic fluxes.

Choose

dynamic control of fluxes’ exponents.

(Correspond to regulation in binding networks)

Choose

stoichiometry-compatible fluxes.

Biological fluxes

Flux exponent

control

Hard limits.

(Control theory tools, e.g. conservation of robustness)

Flux control (FC), incl. FBA

30 of 124

Flux exponent control (FEC) can constrain dynamic fluxes.

Choose

dynamic control of fluxes’ exponents.

(Correspond to regulation in binding networks)

Choose

stoichiometry-compatible fluxes.

Biological fluxes

Flux exponent

control

Flux control (FC), incl. FBA

31 of 124

Flux exponent control (FEC) can constrain dynamic fluxes.

Choose

dynamic control of fluxes’ exponents.

(Correspond to regulation in binding networks)

Choose

stoichiometry-compatible fluxes.

Biological fluxes

Flux exponent

control

e.g. Growth

Flux control (FC), incl. FBA

32 of 124

Flux exponent control (FEC) capture oscillation in glycolysis!

Dano, Sorensen, Hynne,

Nature, 1999

33 of 124

Flux exponent control (FEC) capture oscillation in glycolysis!

Dano, Sorensen, Hynne,

Nature, 1999

34 of 124

FEC captures cell growth arrest under stress.

Cell growth arrest under stress is another hallmark of biological adaptation.

e.g. diauxic growth.

Case study:

Cell grown on glycolysis with sudden increase in maintenance cost.

Jacques Monod, PhD Thesis, 1942.

Jing Shuang (Lisa) Li

(Doyle group)

35 of 124

FEC captures cell growth arrest under stress.

Cell growth arrest under stress is another hallmark of biological adaptation.

e.g. diauxic growth.

Case study:

Cell grown on glycolysis with sudden increase in maintenance cost.

Jacques Monod, PhD Thesis, 1942.

Jing Shuang (Lisa) Li

(Doyle group)

36 of 124

FEC comes from plant-controller splits on

the layered architecture of metabolism.

37 of 124

Layered architecture is shared in control and biology.

38 of 124

Summary

Systems theory

Structures of interaction

FEC enables dynamic constraint-based modeling of metabolism, upgrading state-of-the-art method FBA.

Biocontrol: how cells regulate their behaviors,

viewed as a controllable machine.

Biomolecular reactions

Metabolic machines

Flux exponent control (FEC)

Binding & Catalysis

Tool: Reaction order polyhedra (ROP)

Components

Machines

Components

Machines

Unique application & challenge for MPC!

State and control constraints;

Network sparsity and controller locality;

(localized and distributed MPC via system level synthesis SLS)

Designed controllers ⇔ biological mechanisms.

39 of 124

Funding and collaborators

Collaborators on this project

Funding in part supported by

Army Research Office (ARO) MURI (Contract W911NF-17-1-0402)

This project spans my journey from Caltech to UCSD to Westlake University.

Jing Shuang (Lisa) Li

(Caltech -> Michigan)

John C Doyle

(Caltech CDS)

40 of 124

I thank the inspirations from my mentors.

Erik Winfree

Richard Murray

John Doyle

Rob Phillips

Lior Pachter

Me…

41 of 124

Thank you! Questions?

Imaginations about the

future of biocontrol.

from Vantage Films, by Denis Sibilev, Andrei Myshev, Dmitry Medvedev

42 of 124

To formulate biocontrol, we need theories for components and machines.

Components have measurable properties.

Simply exists.

A purpose or objective.

Machines are functional, specified by architecture.

Systems theory

Structures of interaction

Architecture limits performance.

image: Vecteezy.com

image: Flaticon.com

Components

Machines

Components

Machines

These properties define their interactions.

43 of 124

Examples of interaction structures and systems theories.

Systems theory

Structures of interaction

Mechanical components

Mechanical machine

Lagrangian

Newtonian,

Mass & Force

Components

Machines

Components

Machines

44 of 124

Examples of interaction structures and systems theories.

Gas laws

Transistors, band gap…

Electronics, radiation, amplifier...

Ohm’s law, …

Systems theory

Structures of interaction

Mechanical components

Mechanical machine

Newtonian,

Mass & Force

Steam engine

Thermo-

dynamics

Computer

Turing machine

Electrical circuits

Linear i/o systems

Communication network

Information channels

Components

Machines

Components

Machines

Lagrangian

45 of 124

To formulate biocontrol, we need theories for components and machines.

Systems theory

Structures of interaction

????

Components

Machines

Components

Machines

????

??? Biological components ???

??? Biological machines ???

46 of 124

To formulate biocontrol, we need theories for components and machines.

Systems theory

Structures of interaction

Components

Machines

Components

Machines

Biomolecular reactions

??? Biological machines ???

Binding & Catalysis

Tool: Reaction order polyhedra (ROP)

????

47 of 124

Biomolecular reactions: binding regulates catalysis.

Binding

Catalysis

E. Coli, by David Goodsell.

48 of 124

Biomolecular reactions: binding regulates catalysis.

Binding

Determines the direction of change.

Catalysis

enzyme

substrate

product

total substrate

total product

E. Coli, by David Goodsell.

49 of 124

Biomolecular reactions: binding regulates catalysis.

Binding

Catalysis

enzyme

substrate

product

total substrate

total product

Determines the direction of change.

E. Coli, by David Goodsell.

E. Coli, by David Goodsell.

50 of 124

Biomolecular reactions: binding regulates catalysis.

Binding

Catalysis

enzyme

substrate

product

total substrate

total product

Determines the direction of change.

Determines the rate, regulates catalysis flux.

E. Coli, by David Goodsell.

E. Coli, by David Goodsell.

51 of 124

Biomolecular reactions: binding regulates catalysis.

Binding

Catalysis

enzyme

substrate

product

total substrate

total product

Determines the direction of change.

Determines the rate, regulates catalysis flux.

Slow.

Fast.

52 of 124

Biomolecular reactions: binding regulates catalysis.

Binding

Catalysis

enzyme

substrate

product

total substrate

total product

Determines the direction of change.

Determines the rate, regulates catalysis flux.

Bioregulation

New formulation, not new problem. Only solved under restrictive assumptions.

Slow.

Fast.

53 of 124

Regulation profile of a binding network is hard to solve.

Simplify

54 of 124

Regulation profile of a binding network is hard to solve.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

Totals are variables changed by catalysis.

i.e. as a function of .

Goal: how catalysis flux is regulated.

55 of 124

Regulation profile of a binding network is hard to solve.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

Totals are variables changed in catalysis.

i.e. as a function of .

Goal: how catalysis flux is regulated.

Solve: A degree 2 polynomial.

r=1

r=2

2 s.s. Eqns, ⇒ Degree 3 polynomials.

3 conservations.

56 of 124

Regulation profile of a binding network is hard to solve.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

Totals are variables changed by catalysis.

i.e. as a function of .

Goal: how catalysis flux is regulated.

Solve: Degree 2 polynomial.

n=1

n=2

n

Degree n+1 polynomial…

Degree 3 polynomial.

57 of 124

Regulation profile of a binding network is hard to solve.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

Totals are variables changed in catalysis.

i.e. as a function of .

Goal: how catalysis flux is regulated.

r=r

Degree r+1 polynomials…

Solve: A degree 2 polynomial.

r=1

r=2

Numerical scan: 100 points per dim.

e.g. r=2, dim=4, 100^4 pts....

58 of 124

Regulation profile of a binding network is hard to solve.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

Totals are variables changed in catalysis.

i.e. as a function of .

Goal: how catalysis flux is regulated.

r=r

Degree r+1 polynomials…

Solve: A degree 2 polynomial.

r=1

r=2

Numerical scan: 100 points per dim.

e.g. r=2, dim=4, 100^4 pts....

Intractable analytically or computationally to obtain the full bioregulation profile.

59 of 124

To make progress, people make restrictive assumptions…

Steady state equation:

Conserved quantities:

Assuming

Total substrate:

Total enzyme:

(Michaelis-Menten),

i.e. as a function of .

Goal: how catalysis flux is regulated.

Totals are variables changed in catalysis.

60 of 124

To make progress, people make restrictive assumptions…

Assuming

i.e. as a function of .

Goal: how catalysis flux is regulated.

Combinatorial regulation

Too restrictive for Biocontrol today!!

Highly dynamic

Antebi YE et. al. 2017 Cell

Zhu RH et. al. 2022 Science

Olsman N et. al. 2019 Cell Systems

Briat C. et al 2016 Cell Systems

(Michaelis-Menten),

61 of 124

To make progress, people make restrictive assumptions…

i.e. as a function of .

Goal: how catalysis flux is regulated.

Need to capture the full profile without assumptions.

Rates/fluxes are intractable to solve directly.

Find alternatives to describe bioregulation.

Assuming

(Michaelis-Menten),

62 of 124

Give up rates/fluxes, look at reaction orders.

i.e. as a function of .

Goal: how catalysis flux is regulated.

Assuming

(Michaelis-Menten),

63 of 124

Give up rates/fluxes, look at reaction orders.

i.e. as a function of .

1

0

Two regimes, with 1 or 0 as exponents / reaction orders in tS.

How to capture the full behavior?

Goal: how catalysis flux is regulated.

Assuming

(Michaelis-Menten),

64 of 124

Give up rates/fluxes, look at reaction orders.

i.e. as a function of .

1

0

Log derivative as continuous analog of exponent.

Goal: how catalysis flux is regulated.

Two regimes, with 1 or 0 as exponents / reaction orders in tS.

How to capture the full behavior?

Assuming

(Michaelis-Menten),

65 of 124

Give up rates/fluxes, look at reaction orders.

i.e. as a function of .

1

0

Log derivative as continuous analog of exponent.

Goal: how catalysis flux is regulated.

Two regimes, with 1 or 0 as exponents / reaction orders in tS.

How to capture the full behavior?

Assuming

(Michaelis-Menten),

66 of 124

Reaction order is an alternative representation of bioregulation.

MM formula

0

i.e. as a function of .

Goal: how catalysis flux is regulated.

Assuming

(Michaelis-Menten),

1

67 of 124

Reaction order is an alternative representation of bioregulation.

MM formula

0

i.e. as a function of .

Goal: how catalysis flux is regulated.

Assuming

(Michaelis-Menten),

1

68 of 124

Reaction order is an alternative representation of bioregulation.

(Michaelis-Menten),

MM formula

0

Assuming

i.e. as a function of .

Goal: how catalysis flux is regulated.

Relax this?

1

69 of 124

Reaction order is a HOLISTIC representation of bioregulation.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

Directly use implicit function theorem.

MM formula

70 of 124

Reaction order is a HOLISTIC representation of bioregulation.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

Directly use implicit function theorem.

MM formula

71 of 124

Reaction order is a HOLISTIC representation of bioregulation.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

MM formula

Directly use implicit function theorem.

No assumptions, completely structural.

Directly from binding network topology.

72 of 124

Reaction order is a HOLISTIC representation of bioregulation.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

MM formula

Directly use implicit function theorem.

No assumptions, completely structural.

Directly from binding network topology.

Rates are not solvable.

Approximations are not holistic.

Using reaction orders, we can solve holistically.

73 of 124

Reaction order polyhedra (ROP) capture

full bioregulation profiles.

New holistic foundation, replacing Michaelis-Menten used for 100+ years.

74 of 124

Let’s celebrate this Historic Moment…. of Triangles!

Here we literally have a triangle!

Hydrogen atom of bioregulation!

Rob Phillips

75 of 124

This “hydrogen atom” of bioregulation is an art totem!

Displayed in Chandler cafeteria on Caltech campus.

Activator

Repressor

76 of 124

ROP solves bioregulation problems in full. This is powerful.

Cell fate control and multistability in

Combinatorial regulations

Adaptation and circuit failure in

Highly dynamic scenarios…

Antebi YE et. al. 2017 Cell

Zhu RH et. al. 2022 Science

Olsman N et. al. 2019 Cell Systems

Briat C. et al 2016 Cell Systems

Glycine cleavage system

J Ren, W Wang, J Nie, W Yuan, AP Zeng, 2022

Catalysis with macromolecule substrates.

77 of 124

Example: adaptation to disturbance in reaction order polyhedra.

Degradation

Adapts

Fails to adapt

Fails for high conc. disturbance

78 of 124

Functional regimes are described in reaction order polyhedra.

Repression

Adapts

Fails to adapt

Fails for low conc. disturbance

79 of 124

Both “activating” and “repressing” are present in one binding step.

Activator

Repressor

Substrate acts as an …

80 of 124

Reaction order polyhedra capture full bioregulation profile

(Free enzyme).

81 of 124

Functional regimes are described in reaction order polyhedra.

Degradation

Adapts

Fails to adapt

Fails for high conc. disturbance

82 of 124

TAKE AWAY: Reaction order polyhedra says

Bioregulation MORE THAN saturation/MM/Hill/Monod!

“What are the behaviors for bioregulation (induction curves) in cells?”

Century-old answer: “Michaelis-Menten (MM), of course.”

This is challenged by full bioregulatory profile described by reaction order polyhedra.

  1. Bioregulation is binding’s regulation of catalysis;
  2. Each binding has two extreme scenarios: oversaturation (MM) and tight binding. Beyond single-molecule view.
  3. This gives three archetypal behaviors: Saturation (MM), Limiting factor, hypersensitivity

83 of 124

Reaction order polyhedron captures full bioregulation profile.

MM formula

Overabundant substrate limit.

Exact

84 of 124

Reaction order polyhedron captures full bioregulation profile.

MM formula

Tight binding limit.

Exact

Overabundant substrate limit.

85 of 124

Reaction order polyhedra says

bioregulation >> saturation/MM/Hill/Monod!

86 of 124

Reaction order polyhedra says

bioregulation >> saturation/MM/Hill/Monod!

87 of 124

Reaction order polyhedra says

bioregulation >> saturation/MM/Hill/Monod!

88 of 124

Reaction order polyhedra says

bioregulation >> saturation/MM/Hill/Monod!

89 of 124

Reaction order polyhedra says

bioregulation >> saturation/MM/Hill/Monod!

90 of 124

Bioregulation has THREE archetypal behaviors.

Q: What are the typical behaviors for bioregulation in cells?”

(1) Saturation

(2) Limiting factor

(3) Hypersensitivity

91 of 124

Different archetypal behaviors are cuts through different regimes.

92 of 124

The full bioregulation profile in one binding reaction.

Activator

Repressor

93 of 124

Physical interpretation of reaction orders

94 of 124

Physical interpretation of reaction orders

95 of 124

Reaction order relates internal and external chemical potentials.

Total chemical potential <-> chemical potential of each particle type

96 of 124

Reaction order is a change of coordinates for internal vs external.

97 of 124

Mathematical tools for general applications.

Computational sampling of reaction order polyhedron, applied to induced activator.

98 of 124

Mathematical tools for general applications.

Mathematical technique to directly obtain polyhedron, applied to induced activator.

99 of 124

Mathematical tools for general applications.

Combined to analyze biological behavior holistically, e.g. product inhibition of enzymes.

100 of 124

Mathematical tools for general applications.

Full bioregulation profile reveal hidden adaptive regimes.

101 of 124

Enzyme allostery

102 of 124

Back to our goal: formulate cells as a control system.

1

0

Biomolecular reactions

Binding & Catalysis

Biological machines ???

??

systems theory ??

Mechanical components

Mechanical machine

Lagrangian,

Applied force on mechanical structures

Newtonian,

Mass & Force

Tool: Reaction order polyhedra (ROP)

103 of 124

Back to biocontrol: formulate cells as a controllable machine.

Systems theory

Structures of interaction

Biomolecular reactions

Biological machines ???

Binding & Catalysis

Tool: Reaction order polyhedra (ROP)

????

Components

Machines

Components

Machines

104 of 124

Future work 3: application to systems and synthetic biology

Segall-Shapiro TH et. al. 2018 Nature

Full bioregulation profile reveal hidden adaptive regimes.

105 of 124

Future work 2: dissipative network control for multistability in biomachines

Vertices are special, correspond to monomial birth-death systems.

General behaviors are “convex combination” of vertex behaviors?

General Nonlinear

Biomolecular circuits

Lyapunov or Storage functions:

Quadratic

Entropy-like

Dissipativity: energy expenditure

Dissipativity: reaction order

Easy-to-analyze basis:

106 of 124

Future work 2: dissipative network control for multistability in biomachines

Bistable system.

Any fixed point in grey area has entropy-like Lyapunov function certifying regional stability.

107 of 124

Summary

Systems theory

Structures of interaction

NEXT: Rules of life across scales from biomachine architecture.

Biocontrol: how cells regulate their behaviors,

viewed as a control system.

Biomolecular reactions

Metabolic machines

Flux exponent control (FEC)

Binding & Catalysis

Tool: Reaction order polyhedra (ROP)

Components

Machines

Components

Machines

108 of 124

Overview of the talk

Structural view of biomolecular systems.

Mathematical

Foundation of Structure.

Dynamics from Structure.

Implications on metabolism.

109 of 124

Classical (linear) approach

Structural approach

vs

for (local) system property

System specification

(numerical, exact)

Calculate fixed point and Linearization

(unbounded, sparsity structured)

Determine system property

(parameter specific,

or scan through parameters,

or solve symbolic polynomial inequalities)

System specification

(symbolic, approximate)

Log derivative polytope

(bounded by structure, more than sparsity)

Determine system property

(structural, robust to parameter variations)

110 of 124

Relate structure to fixed point stability.

Derivative:

Positive system

Log derivative:

Additive deviation:

Multiplicative deviation:

Production

Degradation

Constrained structurally.

Varies with rates and conc.

Johan Paulsson et. al.

Michael Savageau et. al.

111 of 124

Relate structure to fixed point stability.

Derivative:

Positive system

Log derivative:

Additive deviation:

Multiplicative deviation:

Production

Degradation

Constrained structurally.

Varies with rates and conc.

Johan Paulsson et. al.

Michael Savageau et. al.

112 of 124

Example: heat shock response in bacteria

Non-saturating regime

From Olsman, Alonso, Doyle 2018

Calculate fixed point and linearization:

Symbolic tests for stability, e.g. Routh Hurwitz rule, need characteristic polynomial:

Log derivative is constant:

All parameters are in time scales!

LMI test:

Any positive fixed pt is structurally stable!

113 of 124

Example: heat shock response in bacteria

Non-saturating regime

Account for saturation effects:

From Olsman, Alonso, Doyle 2018

Calculate fixed point and linearization:

Symbolic tests for stability, e.g. Routh Hurwitz rule, need characteristic polynomial:

Any positive fixed pt, even when saturating, is structurally stable!

114 of 124

From local to global properties.

Vertices are special, correspond to monomial birth-death systems.

General behaviors are “convex combination” of vertex behaviors?

General Nonlinear

Biomolecular circuits

Lyapunov or Storage functions:

Quadratic

Entropy-like

115 of 124

One example, certifying regional multistability.

Bistable system.

Any fixed point in grey area has entropy-like Lyapunov function certifying regional stability.

116 of 124

Vertex of reaction order polyhedra can be scalably computed via zonotopes

117 of 124

Saturation regimes for multistability and oscillations.

Oscillations are hard to describe / analyze.

Repressilator.

No stable fixed point.

Bounded.

2D.

118 of 124

Saturation regimes for multistability and oscillations.

Repressilator:

Stable.

119 of 124

Saturation regimes for multistability and oscillations.

Repressilator:

120 of 124

Saturation regimes for multistability and oscillations.

121 of 124

Log derivative (reaction order) calculations

ALL variables and functions are POSITIVE.

Log vs Linear derivative:

Fold-change

Sums → Weighted average

Terms compete

for DOMINANCE of order

in calculus of positive variables.

Weight

Order

122 of 124

Log derivative (reaction order) calculations

ALL variables and functions are POSITIVE.

Log vs Linear derivative:

Fold-change

Sums → Weighted average

Ratio → Translation

123 of 124

Log derivative operator decomposition

Example:

(Simple binding)

124 of 124

Structure of log derivative comes from stoichiometry.

Steady state equation:

Conserved quantities:

Total substrate:

Total enzyme:

MM formula

Log derivative polytope

Conserved quantities

(log derivative operator decomposition)

Reaction stoichiometry of