Cat
abo
lism
Pre
curs
ors
ATP
Enzymes
Inside every cell
• Autocatalytic feedback (essential)
• Efficient processes
– Minimal enzymes (lean manufacturing)
– Long assembly process (simple steps)
• Limited control feedback
Cat
abo
lism
AA
Ribosome
transl. Proteins
Pre
curs
ors
ATP
ATP
Inside every cell
Ribosomes
make
ribosomes
Translation: Amino acids
polymerized into proteins
Cat
abo
lism
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Building
Blocks
• Translation
• Transcription
• DNA Replication
Cat
abo
lism
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Enzymes
Building
Blocks
Crosslayer
autocatalysis
Macro-layers
Inside every cell
App AppApplications
Router
Cat
abo
lism
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Enzymes
Building
Blocks
Crosslayer
autocatalysis
Macro-layers
Inside every cell
Building
Blocks
Lower layer autocatalysis
Macromolecules making …
Three lower
layers? Yes:
• Translation
• Transcription
• Replication
AA
RibosomeRNA
RNAp
transl. Proteins
xRNAtransc.
Enzymes
DNADNAp
Repl. Gene
AA
RibosomeRNA
RNAp
transl. Proteins
xRNAtransc.
Enzymes
DNADNAp
Repl. Gene
Autocatalytic within lower layers
• Collectively self-replicating
• Ribosomes make ribosomes, etc
Three lower
layers? Yes:
• Translation
• Transcription
• Replication
Naturally
recursive
App AppApplications
Router
Cat
abo
lism
AA
Ribosome
RNA
RNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Enzymes
Building
Blocks
Macro-layers
Naming and addressing
• Names needed to locate objects
• 2.5 ways to resolve a name
1. Exhaustive search, table lookup
2. Name gives hints
• Extra ½ is for indirection
• Address is just a name that involves
locations
Operating systems
• OS allocates and shares diverse
resources among diverse applications
• Clearly separate (disaster otherwise)
– Application name space
– Logical (virtual) name/address space
– Physical (name/) address space
• Name resolution within applications
• Name/address translation across layers
App
kernel
user
In operating systems:
Don’t cross layers
Direct
access to
physical
memory?
Benefits of stricter layering
“Black box” effects of stricter layering
• Portability of applications
• Security of physical address space
• Robustness to application crashes
• Scalability of virtual/real addressing
• Optimization/control by duality?
Bacterial architecture• More complex macro-layering of function
– Upper: Metabolism, envelope, signaling, building blocks
– Lower: Proteins & macromolecule synthesis, replication
• Cleaner layering of control– Transcription factors
– 2 component signal transduction
• Name/address resolution– Global, exhaustive by fast diffusion within layers
– Highly structured interactions between layers
• Limited scalability– Limited to small volumes
– Control proteins scale super-linearly with enzyme
numbers
Reactions
Flow/error
control
Protein
levelAssembly
Flow/error
control
DNA/RNA
levelsInstructions
Building
blocks
1 1
q
h
q Vx
x
1
1y
qk y
yw
xSupply
AA Biosyna
p2Translation
RibosomeTransl.
Production
)(
),( )( tosubj
)( )( max0
P
P
Cx
wcxGR
wVxU l
l
l
i
iix
Does it fit the
framework?
Yes, but it takes
some explaining
and no one has
worked out the
details.
? ?
??
0
0 0
0 0
0 0
( ) subject to max
min ( ) max
min ( ) max
min
Primal:
Dual:
i
i
i
i ix i
i i l li i lp xi l
i i i li l l lp xi l l
p x
U x Rx c
U x p R x c
U x x R p p c
1
( ) max
( ) ( )
i i i i l l
i l
i i i i i i
U x x q p c
U x q x U q
No duality gaps?
Multipath routing?
Coherent pricing?
1 1
q
h
q Vx
x
1
1y
qk y
yw
xSupply
AA Biosyna
p2Translation
RibosomeTransl.
Production
Cat
abo
lism
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Enzymes
Building
Blocks
Crosslayer
autocatalysis
Macro-layers
Cat
abo
lism
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP• Complex machines
− Polymerization
− Complex assembly
• General enzymes
• Regulated recruitment
• Slow, efficient control
• Quantized, digital
• Building blocks
− Scavenge
− Recycle
− Biosynthesis
• Special enzymes
• Allostery, Fast
• Expensive control
• Analog
Cat
abo
lism
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
• DNA replication
− Highly controlled
− Facilitated variation
− Accelerates evolution
• DNA modification (e.g.
methylation)
• Complex RNA control
• Homeostasis
− pH
− Osmolarity
− etc
• Cell envelope
• Movement,
attachment, etc
What we’ve neglected
• Ecosystems
• Biofilms
• Extremophiles
• Pathogens
• Symbiosis
• …
source receiver
control
energymaterials
More
complex
feedback
All these other feedbacks make feedback control
harder, and in each layer biology appears to
cleverly balance competing requirements.
signalinggene expression
metabolismlineage
2 2
0
1ln ln
z z pS j d
z z p
1 1
q
h
q Vx
x
1
1y
qk y
yw
xSupply
AA Biosyna
p2Translation
RibosomeTransl.
Production
Main problem with autocatalytic networks
• Maximize production, but
• Balance risk to fluctuating supply
• (or control for fluctuating demand)
Cat
abo
lism
Pre
curs
ors
ATP
TCA
Gly
G1P
G6P
F6P
F1-6BP
PEP Pyr
Gly3p
13BPG
3PG
2PG
ATP
NADH
Oxa
Cit
ACA
Upper layer autocatalysis
• Fastest allosteric control
• Complex proteins
• High metabolic overhead
• Hard to reprogram
• Essentially analog
Cat
abo
lism
Pre
curs
ors
ATP
TCA
Gly
G1P
G6P
F6P
F1-6BP
PEP Pyr
Gly3p
13BPG
3PG
2PG
ATP
NADH
Oxa
Cit
ACA
Name resolution?
• Locating: Enzymes and
– Substrates
– Allosteric regulator
• Global search by diffusion
• Spatial localization by “solid
state” complexes
PINADNADH
ADPATP
NADPNADPH
CO2 COAACCOA
PPI AMPATP
NH3 AC THFMTH H2S
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
H. Pylori Amino Acid Biosynthesis
S1 S2
S3 S4
ADP
ATP
1
2
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASS
HSE PHS
DHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH
ADPATP
NADPNADPH CO2 COA
ACCOA PPI AMPATP NH3 AC THF
MTH H2S
H Pylori
amino acid
biosynthesis
As a bipartite
labeled graph.
Substrates
Carriers
1
2
3
4
1 0
1 0
0 1
0 1
1 1
1 1
S
S
S
S
ATP
ADP
H Pylori
amino acid
biosynthesis
As a color coded
(for reversibility)
stoichiometry
matrix.
21
12
23
50
61
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASS
HSE PHS
DHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH
ADPATP
NADPNADPH CO2 COA
ACCOA PPI AMPATP NH3 AC THF
MTH H2S 21
12
23
50
61
S1 S2
S3 S4
ADP
ATP
1
2
Substrates
Carriers
1
2
3
4
1 0
1 0
0 1
0 1
1 1
1 1
S
S
S
S
ATP
ADP
These are
equivalent to each
other but not to
unipartite graphs.
1 2
3 4
S ATP S ADP
S ATP S ADP
S1 S2
S3 S4
ADP
ATP
1
2
S1 S2
S3 S4
ADP
ATP
Substrate graph
1
2
Reaction graph
Unipartite projections
lose too much.
S1 S2
S3 S4
ADP
ATP
1
2
S1 S2
S3 S4
ADP
ATP
Substrate graphSuppose these reactions
are in different modules,
say,
Lipid
biosyn
AA
biosyn
S2
S4
“Small world?”
Not really.
PINADNADH
ADPATP
NADPNADPH
CO2 COAACCOA
PPI AMPATP
NH3 AC THFMTH H2S
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
H. Pylori Amino Acid Biosynthesis
ADPATP
NADPNADPH
DHS SME S5PDQT DHS
“Typical” reactions
PINADNADH
ADPATP
NADPNADPH
CO2 COAACCOA
PPI AMPATP
NH3 AC THFMTH H2S
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
ADPATP
NADPNADPH
DHS SME S5PDQT DHS
“Typical” reactions
PINADNADH
ADPATP
NADPNADPH
CO2 COAACCOA
PPI AMPATP
NH3 AC THFMTH H2S
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
precursors amino acids
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
precursors amino acids
PINADNADH
ADPATP
NADPNADPH
CO2 COAACCOA
PPI AMPATP
NH3 AC THFMTH H2S
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
Aside
• A popular view of “modularity” is
– High connectivity within the module
– Low connectivity to the outside
• This is intuitively appealing, and there are some
examples…
• …but the most important elements of biological
modularity are often exactly the opposite of this
PINADNADH
ADPATP
NADPNADPH
CO2
Highest degree carriers
PINADNADH
ADPATP
NADPNADPH
CO2
Highest degree carriers
The carriers are a crucial
element of modularity,
But they don’t have “high
internal, low external”
connectivity.
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
DAH
PHP
AKG
precursors
The precursors are
a crucial element of
modularity,
But they don’t have
“high internal, low
external”
connectivity.
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASS
HSE PHS
DHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
Without carriers
Long assembly lines
“long” not “small” worlds
21
12
23
50
61
carriers
metabolites
reactions “Vertical”
decomposition
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASS
HSE PHS
DHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH
ADPATP
NADPNADPH
CO2 COAACCOA
PPI AMPATP
NH3 AC THFMTH
H2S
21
12
23
50
61
carriers
metabolites
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASS
HSE PHS
DHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH
ADPATP
NADPNADPH
CO2 COAACCOA
PPI AMPATP
NH3 AC THFMTH
H2S
Prices?
• Each constrained quantity has a carrier
• Delivery by rapid diffusion
• “Price” by concentration of charged carrier?
• Elegant implementation of optimization and
duality, integrated with delivery?
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
• Fastest allosteric feedback control
• Complex proteins
• High metabolic overhead
• Hard to reprogram
1( )k ATP
x
1( )k x y
x
Control
Autocatalytic
• Fastest allosteric feedback control
• Complex proteins• High metabolic overhead
• Hard to reprogram
1( )k x y
xAutocatalytic
Control
1( )k x
x
level
form/activity
rate
Layered control
• How to get rid of the RHP zero?
• What are the new tradeoffs?
y
x
More
control
y
x
More
complex
enzymes
y
x
Biology appears
to do both
AA Ribosometransl.
Lower layer autocatalysis
Ribosomes making ribosomes
AA
Ribosome
RNA
RNAp
transl.Proteins
xRNAtransc.
DNADNAp
Repl. Gene
ATP
Lower layer autocatalysis
Macromolecules making …
AA
RibosomeRNA
RNAp
transl. Proteins
xRNAtransc.
Enzymes
DNADNAp
Repl. Gene
Autocatalytic within lower layers
• Collectively self-replicating
• Ribosomes make ribosomes, etc
Three lower
layers? Yes:
• Translation
• Transcription
• Replication
Naturally
recursive
Flow/error
Reactions
Protein level
Flow/error
Translation
RNA level
Flow/error
Transcription
DNA level
Naturally
recursive
Protein level
Flow/error
Translation
RNA level
Flow/error
Transcription
DNA level
Three lower
layers? Yes:
• Translation
• Transcription
• Replication/
rearrangement
Replication
DNA Replication/
Rearrangement is
complex and
highly controlled Flow/error
Cat
abo
lism
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Enzymes
Building
Blocks
Crosslayer autocatalysis
Supply/demand control between layers?
Reactions
Protein Assembly
DNA/RNA
control
control
DNA
Ou
tsid
e
Ins
ide
Tra
nsm
itte
r
Ligands &
Receptors
Rec
eiver
Responses
control
Receiver
Resp
on
ses
Tra
nsm
itte
r
Rec
eiverVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiverVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiverVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiverVariety of
Ligands &
Receptors
Variety of
responses
• 50 such “two component” systems in E. Coli
• All use the same protocol
- Histidine autokinase transmitter
- Aspartyl phospho-acceptor receiver
• Huge variety of receptors and responses
• Also multistage (phosphorelay) versions
Signal
transduction
Tra
nsm
itte
r
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiverVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiv
erVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiverVariety of
Ligands &
Receptors
Variety of
responsesT
ran
smit
ter
Rec
eiverVariety of
Ligands &
Receptors
Variety of
responses
Tra
nsm
itte
r
Rec
eiver
Ligands &
ReceptorsResponses
Shared
protocols
Flow of “signal”
Recognition,
specificity
• “Name resolution” within signal transduction
• Transmitter must locate “cognate” receiver
and avoid non-cognate receivers
• Global search by rapid, local diffusion
• Limited to very small volumes
Tra
nsm
itte
r
Rec
eiver
Ligands &
ReceptorsResponses
Tra
nsm
itte
r
Rec
eiver
Ligands &
ReceptorsResponses
Tra
nsm
itte
r
Rec
eiver
Ligands &
ReceptorsResponses
Shared
protocols
Flow of “signal”
Recognition,
specificity
Variety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
Ligands &
Receptors
Variety of
Ligands &
Receptors
Variety of
Ligands &
Receptors
Variety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
responsesVariety of
Ligands &
Receptors
Variety of
responses
Tra
nsm
itte
r
Rec
eiver
Tra
nsm
itte
r
Rec
eiver
Tra
nsm
itte
r
Rec
eiver
Recognition,
specificity
Huge variety• Combinatorial
• Almost digital
• Easily reprogrammed
• Located by diffusion
Tra
nsm
itte
r
Rec
eiver
Tra
nsm
itte
r
Rec
eiver
Tra
nsm
itte
r
Rec
eiver
Flow of “signal”
Limited variety• Fast, analog (via #)
• Hard to change
Reusable in
different pathways
Internet
sitesUsers
Tra
nsm
itte
r
Rec
eiver
Ligands &
ReceptorsResponses
Shared
protocols
Flow of “signal”
Recognition,
specificity
Note: Any
wireless system
and the Internet
to which it is
connected work
the same way.
Lap
top
Rou
ter
Flow of packets
Recognition,
Specificity (MAC)
Tra
nsm
itte
r
Ligands &
Receptors
Rec
eiver
Responses
“Name” recognition
= molecular recognition
= localized functionally
= global spatially
Transcription factors
do “name” to “address”
translation
DNA
Tra
nsm
itte
r
Ligands &
Receptors
Rec
eiver
Responses
“Addressing”
= molecular recognition
= localized spatiallyBoth are
• Almost digital
• Highly
programmable
“Name” recognition
= molecular recognition
= localized functionally
Transcription factors
do “name” to “address”
translation
DNA
2CST systems provide
speed, flexibility,
external sensing,
computation, impedance
match, more feedback,
but
greater complexity and
overhead
Tra
nsm
itte
r
Ligands &
Receptors
Rec
eiver
Responses
ReceiverR
espon
ses
Feedback control
There are simpler
transcription
factors for sensing
internal states
DNA
There are simpler
transcription
factors for sensing
internal states
DNA
RNAp
DNA and RNAp
binding domains
Sensor domains
Domains can
be evolved
independently
or coordinated.
Application
layer cannot
access DNA
directly.Highly
evolvable
architecture.
There are simpler
transcription
factors for sensing
internal states
DNA
RNAp
DNA and RNAp
binding domains
Sensor domainsThis is like a
“name to
address”
translation.
Sensing the
demand of the
application
layer
Initiating
the change
in supply
Sensor
domains
Sensing the
demand of the
application
layer
Any
input
Any
other
input
• Sensor sides attach to metabolites or other proteins
• This causes an allosteric (shape) change
• (Sensing is largely analog (# of bound proteins))
• Effecting the DNA/RNAp binding domains
• Protein and DNA/RNAp recognition is more digital
• Extensively discussed in both Ptashne and Alon
DNA and RNAp
binding domains
DNA and RNAp
binding domains
Sensor
domains
“Cross talk” can be
finely controlled
Any
input
Any
other
input
• Application layer signals can be integrated or not
• Huge combinatorial space of (mis)matching shapes
• A functionally meaningful “name space”
• Highly adaptable architecture
• Interactions are fast (but expensive)
• Return to this issue in “signal transduction”
DNA
“Addressing”
= molecular recognition
= localized spatially
Both are
• Almost digital
• Highly
programmable
“Name” recognition
= molecular recognition
= localized functionally
= global spatially
Transcription factors
do “name” to “address”
translation
RNAp
Can activate
or repress
And work in
complex logical
combinations
Promoter Gene1 Gene2 …
• Both protein and DNA sides have sequence/shape
• Huge combinatorial space of “addresses”
• Modest amount of “logic” can be done at promoter
• Transcription is very noise (but efficient)
• Extremely adaptable architecture
Promoter Gene5 Gene6 …
(almost analog)
rate determined
by relative copy
number Binding
recognition
nearly digital
Promoter Gene5 Gene6 …
rate (almost analog)
determined by copy number
Recall: can work by
pulse code
modulation so for
small copy number
does digital to
analog conversion
Promoter Gene1 Gene2 …
No crossing layers• Highly structured interactions
• Transcription factor proteins
control all cross-layer interactions
• DNA layer details hidden from
application layer
• Robust and evolvable
• Functional (and global) demand
mapped logically to local supply
chain processes
Any
input
Reactions
Assembly
DNA/RNA
control
control
DNA
Ou
tsid
e
Ins
ide
Tra
nsm
itte
r
Ligands &
Receptors
Rec
eiver
Responses
control
Receiver
Resp
on
ses
Protein
Cross-layer control
• Highly organized
• Prices? Duality?
• Minimal case study?
Reactions
Flow/error
control
Protein
levelAssembly
Flow/error
control
DNA/RNA
levelsInstructions
Building
blocks
0
0 0
0 0
0 0
( ) subject to max
min ( ) max
min ( ) max
min
Primal:
Dual:
i
i
i
i ix i
i i l li i lp xi l
i i i li l l lp xi l l
p x
U x Rx c
U x p R x c
U x x R p p c
1
( ) max
( ) ( )
i i i i l l
i l
i i i i i i
U x x q p c
U x q x U q
No duality gaps?
Multipath routing?
Coherent pricing?
1 1
q
h
q Vx
x
1
1y
qk y
yw
xSupply
AA Biosyna
p2Translation
RibosomeTransl.
Production
Any
input
GLC
DPG
PGLPGC RL5P
X5P6PG
MALCIT
ICIT
SUCFUM
G6P
F6P
T3P
3PGPEP
R5P
E4P
PYROA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHOAN NAN CD5 IGP
PPN HPP
BAPASSHSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPSASE
GLN
SER
TRP
ASP
TYR
THRLYS
CYS
GLYASN
AKG GLU
RNA
Gene
Transc.
RNAp
trans. EnzymesAA
mRNA
• Slowest transcription control
• Complex transcription factors
• Lowest metabolic overhead
• Easily reprogrammed
All transcriptional
regulatory links
are downward.
Nodes are
operons. Global
regulators are
red. Yellow
marked nodes are
operons in the
longest regulatory
pathway related
with flagella
motility.Ma et al. BMC
Bioinformatics 2004
5:199 doi:10.1186/1471-
2105-5-199
Hierarchical
structure of
E. coli transc.
regulatory
network.
Note: all feedback in this picture has been
removed in two ways:
1) There are self-loops
where an operon is
controlled by one it’s
own genes
2) All the real complex
control is in the
protein interactions
not shown (e.g. see
heat shock details)
These are not really
control systems,
they just initiate
manufacturing
This architecture has limited scalability:
1) Fast diffusion can
only work in small
volumes
2) The number of
proteins required
for control grows
superlinearly with
the number of
enzymes (Mattick)
“Control”
Enzymes
Any mRNA
Translation
Initiation codon
Any
inputAny
protein
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
AKG GLU
• Fast translation control
• Complex RNAs
• Medium metabolic overhead
• Highly reprogrammable?
GLC
DPG
PGLPGC RL5P
X5P6PG
MALCIT
ICIT
SUCFUM
G6P
F6P
T3P
3PGPEP
R5P
E4P
PYROA
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHOAN NAN CD5 IGP
PPN HPP
BAPASSHSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPSASE
GLN
SER
TRP
ASP
TYR
THRLYS
CYS
GLYASN
AKG GLU
Any
input
RNA
Gene
Transc.
RNAp
trans. EnzymesAA
mRNA
• Slowest transcription control
• Complex transcription factors
• Lowest metabolic overhead
• Easily reprogrammed
Any mRNA
Translation
Initiation codon
Any
inputAny
protein
• Fast translation control
• Complex RNAs
• Medium metabolic overhead
• Highly reprogrammable?
• Fastest allosteric feedback control
• Complex proteins
• High metabolic overhead
• Hard to reprogram
App AppApplications
Router
Clean slate layering?• Two “macrolayers” with a new, higher “waist”
– Upper: Managing content, function, naming
– Lower: Managing physical resources, addressing
• Lower layers: map to physical addresses (PNA)
– Recursive “microlayers” of control and management
– Different scopes (more global and lumped to more
local and detailed)
– No global addresses, hide details, addresses
• Cleaner role of optimization and control?
• Integration with naming and addressing
• Align robustness and security
SlowFast
Wasteful
Efficient
log
log
Example design
space:
Speed versus
efficiency
Faster
Cheaper
Design tradeoffs
waste
ful
fragile
Sharpen
hard
bounds
bad
Find and
fix bugs
Complementary
approaches
• Each focuses on few dimensions
• Important tradeoffs are across these dimensions
• Speed vs efficiency vs robustness vs …
• Robustness is most important for complexity
• Need “clean slate” theories
• Progress is encouraging
waste
ful
fragile?
slow
?
• Thermodynamics (Carnot)
• Communications (Shannon)
• Control (Bode)
• Computation (Turing)
Standard theories are severely limited
Robust
• Secure
• Scalable
• Evolvable
• Verifiable
• Maintainable
• Designable
• …
Fragile
• Not …
• Unverifiable
• Frozen
•…
Most dimensions are robustnessCollapse for visualization
fragile
wasteful
fragile
• Important tradeoffs are across these dimensions
• Speed vs efficiency vsrobustness vs …
• Robustness is most important for complexity
• Collapse efficiency dimensions
waste
resources
waste time
log
log
bad
But many existing
systems and architectures
are clearly far from any
fundamental limits.
?
??
So fixing “bugs”
in existing
architectures
has most
immediate
impact.
waste
ful
fragile
Note: “log” suggests
orders of magnitude
variations
Wasteful
Efficient
log
log
Brains?
Cells?
Conjecture: Cells and
brains are RYF but not
gratuitously fragile
fragile
They avoid
cross-layering?
SlowFast
Wasteful
Efficient
log
log
DNA
Neurons
CMOS
Brains
What makes this possible?
Network
architecture
Cells
Embedded
virtual
actuator/
sensor
Network
cable
Controller
Lib
App
DIF
Networked embedded
Lib
Physical
plant
Embedded
virtual
actuator/
sensor
Network
cable
Controller
DIF
Physical
plant
Meta-layering of cyber-phys control
Physiology
Organs
Cells
Meta-layers
Physiology
Organs
Meta-layers
Prediction
Goals
Actions
errors
ActionsCo
rte
x
Physiology
Organs
Meta-layers
Prediction
Goals
Actions
errors
ActionsFrom
Information to
“Outformation” to
“Actformation”?
Physiology
Organs
Cells
Meta-layers
Maximize
allowable
fluctuations
in
Minimize
resulting
fluctuations
in(Evolution +
physiology)
2SpO
BP
wattsSimple
starting point.
2SpO
BP
Maximize
MinimizeEV
HR
Control
error
error
Control
requirement
watts
functional requirements
Result of
control
requirements
2SpO
BP
EV
HR2VO
2VCO
watts
error
error
low
varhigh
watts
Control
requirement
Control
requirement
Finally VO2 and VCO2 don’t need tight
control and vary as needed, they don’t
change as much as watts, but much
more than spO2 or BP.
