Press ? or Esc to close.
RFCs from Genetic Assemblies. Short technical documents on improving the fundamentals of biotechnology in order to run more experiments cheaper.
Written in 80 column plaintext, as the creators intended.
Genetic Assemblies RFC 0
Genetic Assemblies Technology Corporation (GATC)
Request For Comments (RFC)
Keoni Gandall
<keoni@geneticassembies.com>
29 Sep 2025
The founding vision of synthetic biology was that genetic parts could be
defined and reused, like electrical components. This concept led to the
BioBricks RFCs (Request For Comments), where text documents described protocols
which could define genetic parts so that they could be reused. Unfortunately,
it turned out genetic parts were difficult to reuse, and thus the field moved
in alternative directions.
I believe the most impactful direction of synthetic biology should be towards
lowering the cost, difficulty, and accessibility of running biological
experiments. We will do this by changing the most fundamental operations of
biology so that they are either cheaper or easier to automate.
GATC RFCs are a new generation of RFCs by Keoni Gandall and Genetic Assemblies
Technology Corporation (GATC) on specific methods that can be used to make
biological protocols cheaper and easier to automate. By adopting these methods
and advocating for other organizations to adopt them as well, we hope to
increase the amount of data we can collect about the biological world and the
number of people who can create new, beautiful biotechnological ideas.
- Keoni Gandall
Genetic Assemblies RFC 1
Acinetobacter baylyi for Cloning Plasmids
<Proposal>
Keoni Gandall
<keoni@geneticassemblies.com>
02 Oct 2025
0. Introduction
We propose Acinetobacter baylyi as a better alternative for basic plasmid
cloning than Escherichia coli or Vibrio natriegens. A. baylyi is a
gram-negative bacterium with a constitutively active natural competence system.
This is in contrast to other organisms, such as E. coli or V. natriegens, which
require special competence inducing protocols, which usually require
temperature controlled conditions, bulk cell manipulation, or specialized
electronics.
A. baylyi solves these problems through its natural competence system, which
actively takes up exogenous DNA during exponential growth. It is fast enough to
be practical, requires almost no preparation for transformations, and
competence is not dependent upon cold conditions. These properties make it
simple enough to use for inexperienced users, quick to use for advanced users,
and most importantly, enables full automation of transformations without
expensive hardware, such as automated freezers and cold storage devices.
1. Background
A. baylyi is a non-motile, strictly aerobic gram-negative bacterium with a ~35
minute doubling time[1]. It was originally derived from soil samples and is
known to have diverse metabolic pathways for breaking down aromatic compounds.
It does not have the ability to do fermentation and uses the Entner-Doudoroff
(ED) Pathway rather than the Embden-Meyerhof-Parnas (EMP) Pathway, making it
relatively less efficient at digesting sugar than other organisms like
Escherichia coli or Vibrio natriegens.
The most remarkable ability of A. baylyi is that it constitutively produces
natural competence machinery, and thus takes up DNA from the environment during
exponential growth. This is unlike the natural competence machinery of Bacillus
subtilis or Vibrio natriegens, which is only produced in specific starvation
conditions.
2. Practical use
We choose A. baylyi as a target strain because the fundamentals of its biology
make it compatible with simpler hardware. Its natural competence system has been
proven to be efficient enough to do multi-part GoldenGate assemblies[2] while
still only requiring liquid-shaking incubation, which is a requirement for
growing strains anyway. Since this is based on growth, it does not require cold
storage or centrifugation to create competent cells or do transformations.
On the most fundamental level, A. baylyi may not be the fastest, nor the most
used, nor have the most interesting metabolism of any organism, but it does one
thing better than any other organism: uptake DNA from the environment. The goal
of cloning is to get DNA that we synthesize or assemble into an organism, and at
that specific task, A. baylyi is the best.
Growth is 30c at typical bacterial media (LB). Unlike commonly used bacteria, A.
baylyi must be grown in O2 rich conditions, as it cannot undergo fermentation.
3. Comparison to other organisms
1. Escherichia coli competent cell prep requires, at minimum, cold storage
conditions and a centrifuge. This makes the preparation of competent cells
on a robotic deck difficult. The requirement for freezing competent cells
means for an automated protocol, either a human needs to load the deck, or
automated -80c freezer integration is necessary (which is expensive and
difficult).
2. Vibrio natriegens competent cells either are equally complex as E. coli
competent cell preps, or use natural competence. However, natural
competence preparations of V. natriegens require starvation conditions and
their total efficiency is orders of magnitude lower than A. baylyi.
4. Limitations
A. baylyi does have some limitations:
1. Common E. coli origin of replications (oris) do not function in A. baylyi
(ColE1, p15a, SC101, etc), so new oris are needed. Of the currently proven
oris, there are none which are both high copy and stable.
2. Many common antibiotic resistance markers do not work well (e.g.
ampicillin) and A. baylyi can rapidly form spontaneous resistance to many
antibiotics (chloramphenicol).
3. Natural competence can lead to failure modes which have not been
thoroughly explored. We do not know how or where it fails in comparison to
typical E. coli culture.
4. A. baylyi does not grow as quickly as E. coli and nowhere as quickly as V.
natriegens. (Author's note: empirically, it seems to grow approximately as
fast on solid agar media as E. coli).
5. Engineering tools
5.1. Origin of Replications
There are a couple origin of replications that function well in A. baylyi. The
plasmid pBAV1K-T5-gfp[3] is the only high copy origin, but is known to be
relatively unstable. RSF1010 also works[4] and is reported as more stable than
pBAV1K, albeit at a lower copy number. pWH1266, a native Acinetobacter plasmid
origin, is also known to function, but may affect adjacent promoters[5].
5.2. Resistance Markers
Antibiotics are more tricky in A. baylyi than E. coli. This is because A. baylyi
is a soil-associated bacterium, and so there are greater competitive pressures
with nearby fungi and bacteria. In part, the ability for A. baylyi to take up
genes from its environment means it has an arsenal of generic antibiotic pumps
which can be upregulated to cause spontaneous resistance[6].
The only antibiotics that are known to work without spontaneous resistance are
kanamycin (and kanamycin family, like gentamycin) and tetracycline. The genes
for spontaneous resistance of chloramphenicol are known[6], and there is the
potential for these to be engineered to allow for robust chloramphenicol
selection.
5.3. Negative selection
tdk is a proven negative selection gene in A. baylyi. It encodes thymidine
kinase which, in the presence of AZT (3'-azido-2',3'-dideoxythymidine or
zidovudine), a thymidine analog, causes chain termination during DNA synthesis.
This kills the cells, and provides a method of negative selection. It is much
cheaper to use zidovudine, the pharmaceutical name of AZT, for negative
selection, and it works just as well.
5.4. Transformation
Transformation is done by mixing A. baylyi with fresh media and growing in the
presence of the target DNA, and this process is flexible. The initial seed
culture can be directly from a glycerol stock, from an overnight culture, or
from a plate. An efficient protocol is as follows: 250uL of LB, 17.5uL of
overnight culture, and ~1 to ~10uL of target DNA (this can be directly from an
assembly reaction). Grow for 3hr at 30c, shaking, and then plate.
6. Next Steps
The requirement of alternative origins and resistance markers makes the switch
from E. coli to A. baylyi difficult, but we believe it is a worthwhile endeavor.
The next step will be to create a vector backbone that is both high copy and
stable, and to begin integrating that backbone into common cloning pathways. We
are working on that right now.
Strains and plasmids available upon request. Please join us in lowering the cost
and effort of cloning by using fundamentally better biology.
- Keoni Gandall
X. References
1. Reduced Mutation Rate and Increased Transformability of Transposon-Free
Acinetobacter baylyi ADP1-ISx
2. Rapid and assured genetic engineering methods applied to Acinetobacter
baylyi ADP1 genome streamlining
3. Rational Design of a Plasmid Origin That Replicates Efficiently in Both
Gram-Positive and Gram-Negative Bacteria
4. Development of a genetic toolset for the highly engineerable and
metabolically versatile Acinetobacter baylyi ADP1
5. Expression Vectors for Acinetobacter baylyi ADP1
6. Single-Step Selection of Drug Resistant Acinetobacter baylyi ADP1 Mutants
Reveals a Functional Redundancy in the Recruitment of Multidrug Efflux
Systems
Genetic Assemblies RFC 2
Implementation of Simple Bioautomation
<Opinion>
Keoni Gandall
<keoni@geneticassemblies.com>
05 Oct 2025
0. Introduction
There is an inherent tradeoff between simplicity of interface and simplicity of
implementation. In the seminal essay "The Rise of Worse is Better" (1989), the
author, Richard P. Gabriel, argues that simplicity of implementation is more
important than simplicity in interface, because simple implementations could be
immediately used and spread, then adapting and surviving for their environment.
This was a keen observation for 1989. But things changed. And python is proof.
Python took the opposite approach: its implementation is anything but simple,
but it took simple interfaces to the extreme, and through this, won. As of
writing, python accounts for more than 1/4 of all code written in all languages,
and, with AI only accelerating the writing of python. Was Richard simply wrong?
Richard wasn't wrong, he was just looking backwards instead of forwards. In
scarce environments where hardware is expensive and labor is cheap, like
computer science in the 80s, simplicity of implementation allows for rapid
spread. Users could take the time to learn, while simple implementation allowed
for more rapid iteration in portability and performance, which mattered when
hardware was expensive. This enabled viral spread of simply implemented
software.
On the other hand, python began to win when these factors flipped: hardware
exponentially decreased in cost while the rise of the internet put software
developers in massive demand. In these conditions, simplicity of interface, not
implementation, enabled viral spread, as user's time was expensive and machines
to run that software was cheap.
It is important to note that the simplicity of interface is often built on top
of simplicity of implementation. The most popular python libraries are built on
C - the quintessential "simple implementation" language. This is also true in
other areas: LLMs can be implemented in only a few thousand lines of code, and
this property was vital in their emergence (as one could scale training), while
what made them spread is a simple text interface.
1. Spreading Bioautomation
The concept of simple implementations and simple interfaces can be applied to
biotechnology and bioautomation. Someday, when hardware and capabilities are
abundant, simple interfaces will win. But that day is not today: what matters
today is that the composable implementation of machines towards biology
experiments works, in some capacity, and can be spread.
The key is the "spread". There needs to be some kind of viral function, just
like how Unix and C spread. Automating drug discovery is not viral, in fact, it
is anti-viral: the pharma companies working with you explicitly do not want to
share how they are doing things. There needs to be a way to work with machines
to complete biological experiments that encourages its own spread, not works
against its own spread for "competitive advantage".
Robotics companies are also not incentivized to do so. If you spread a
simply-implemented system that works, you immediately open yourself up to
commoditization. This is exactly what happened to IBM vs Compaq in the 80s - and
it is likely to happen again. By making portable software that works, people
port it to cheaper platforms. Opentrons had a brief period of time where they
could have started this (as they had the commoditized hardware), but since there
wasn't a portable system, people couldn't easily port their protocols. While
this is a competitive advantage for any individual hardware company, it does
prevent acceleration in our field.
2. Inventory Management
For a simply-implemented system, there can be tons of tacit knowledge necessary
to run the robots: it has just got to work. It can be ported to other robots
later. What prevents basic robots, like the Opentrons OT2 or Hamilton STAR from
having rich libraries of protocols and people building on those? The answer is
inventory management.
Robotic systems are set up differently with different hardware. Some have
certain pipettes, some have certain heater-shakers, others have different
incubator integrations. Even worse than a variety of hardware, each lab has
different reagents sources, different ways of creating media, and different cell
lines.
Even with the exact same robots, there are hardware differences and wetware
differences. Among different labs there are also software differences. And
because of all of these differences, producing shareable protocols is nearly
impossible.
3. Cloud Lab
The obvious solution is to have a single source of hardware (ie, people share
the hardware) and a centralized inventory system, so that people can share
strains and genetic material for their protocols. These two combined, with
minimally viable software to run the whole thing, becomes very attractive (and
of course, it actually has to run simple experiments).
Of all the cloud labs that have been built, none have actually attempted this
kind of sharing-first system. There have been none where reagents are simple and
shareable. There haven't been any that cheaply run simple experiments (mostly
because of over-capitalization: as they raised for the hardware, they can only
sell to pharma). There is too much focus on software and not enough on the truly
scarce resource that a cloud lab enables: instant sharing.
It should be obvious that instant sharing and communication would be important:
look at how the internet made computers explode in popularity. Right now, to
share materials, universities and companies have to sign lengthy MTAs and wait
weeks for physical shipments. What if you could instantly access the strains and
DNA you needed for experiments? What if you knew you could get the hardware to
reproduce an experiment within a few minutes instead of negotiating with a robot
supplier for months?
The interface can be simple and frankly, kind of garbage. But it must allow for
creativity, it must actually run experiments, and things must be shareable. That
is how we move people from doing experiments by hand to doing experiments on
robots and enable new beautiful biology to be built by anyone.
- Keoni Gandall
Genetic Assemblies RFC 3
Conjugation as Primary Transfection Mechanism
<Proposal>
Keoni Gandall
<keoni@geneticassemblies.com>
28 September 2025
Introduction
We propose conjugation as the primary transfection method for transforming non-
cloning strains. While this approach cannot effectively transform mammalian cell
lines, a wide range of bacterial, archaeal, fungal, and plant cell types can be
transformed through conjugation in a standardizable way, without cell-specific
competence induction protocols. In contrast to specialized transformation
protocols for each cell type, conjugation enables standardized protocols for a
diverse range of different strains, making the entire process of engineering
novel organisms more amenable to automation.
0. Background
Bacterial conjugation is the process in which a bacterial cell transfers genetic
material to another cell through the use of a genetically encoded pilus. This
process usually occurs between bacteria, but can occur between domains to
plants[1], fungi[2], animals (human cells)[3], and archaea[4]. A remarkable
example is Agrobacterium, which can transform tree cells into a tumor-like gall
that produces and secretes nopaline, a nitrogen source that only Agrobacterium
can use. Agrobacterium-based transformation has gone on to be one of the main
methods that scientists engineer plant cells.
Promiscuous conjugation systems, like IncP (Escherichia coli) or Ti
(Agrobacterium), typically mate with cells promiscuously by proximity. By
growing cells together either in liquid culture or, more efficiently, on solid
media, DNA can be transferred to nearly any organism.
Importantly, this transfer can be directly done from bacteria cells used for
cloning to target cell types. It does not require refigeration, centrifugation,
or electroporation, and does not require specialized chemicals for each target
organism. Instead, it enables simple hardware and basic chemicals to do
transformations of a wide variety of cells.
1. Implementation
The implementation is simple: have the IncP conjugation machinery in a gram
negative cell. This can either be through an IncP plasmid, like pRK24 [5], or a
conjugation strain, like S17. Moving to other organisms like Acinetobacter
baylyi or Vibrio natriegens is theoretically also possible, but untested.
Each organism <-> organism will require specialized protocols, and are beyond
the scope of this specific RFC. Rather, this RFC is intended to be a proposal
for a fundamentally different way to think about how we could transform
organisms versus what we do today, and to be referenced back to later with
specific implementations.
The volumes and temperatures involved make conjugation compatible with simple
automated systems: strains can be grown and handled in ways that don't require
human intervention. While sometimes electroporation may be necessary (either
for transformation efficiency for library preps or recalcitrant strains), basic
everyday flows between cloning strains and target strains can be made far more
efficient than they are right now.
The ramifications of opening up far more strains for low-cost and reliable
engineering are predictable. And good!
- Keoni Gandall
X. References
1. Coordinated regulation of octopine degradation and conjugative transfer of Ti plasmids in A. tumefaciens
2. A Fast and Practical Yeast Transformation Method Mediated by Escherichia coli Based on a Trans-Kingdom Conjugal Transfer System: Just Mix Two Cultures and Wait One Hour
3. Genetic transformation of HeLa cells by Agrobacterium
4. Interdomain Conjugal Transfer of DNA from Bacteria to Archaea
5. https://www.addgene.org/51950/
Genetic Assemblies RFC 4
Porting IncP conjugation to Acinetobacter baylyi
<Proposal><WorkInProgress>
Keoni Gandall
<keoni@geneticassemblies.com>
28 September 2025
Introduction
We propose porting the IncP conjugation machinery (RP4 plasmid, S17 strain)
from Escherichia coli to Acinetobacter baylyi in order to reap the benefits of
both Genetic Assemblies RFC 1 and RFC 3. In RFC 1, we propose the use of A.
baylyi for transformation because it efficiently takes up DNA from the
environment without special preparation. In RFC 3, we propose the use of
conjugation as the primary method to transfer DNA into alternative cell types.
By combining both, we uncover a method to shuttle between DNA in a tube and
almost any organism of interest through extremely simple mixing procedures.
This both increases accessibility of many organisms to engineering by general
users and facilitates automated engineering of those organisms.
0. Background
We propose to port the transformation machinery for the incP conjugative pilus and
transfer system to A. baylyi. The easiest method is using the pRK24 plasmid: We
can simply directly conjugate this plasmid into A. baylyi and select with
tetracycline, which is one of the antibiotics which works well in that strain.
Before we have a specialized strain, this will have to do.
However, we have more ambitious plans for A. baylyi. A plasmid system is
disadvantageous for several reasons: it uses one of the few antibiotics that
work well against A. baylyi, it is difficult to negatively select against, it
can be transferred alongside the target DNA, and you must maintain selection
lest it be evolved out of the population.
Instead, we are aiming at porting the necessary components to be genomically
encoded. These can be stably maintained without antibiotic selection, have no
danger of being transferred along with the target DNA, and allow for 1 simple
strain to be used for all preparations.
In addition, we aim to knock out alr/dadX to remove the ability of A. baylyi to
grow without D-alanine supplementation in rich media. D-alanine is used in the
bacterial cell wall, so is a necessary nutrient, but rich media does not supply
it (as eukaryotic cells do not have need for much D-alanine, so it is not in the
yeast extract or tryptone). This allows us to mate into target strains, then
remove the donor A. baylyi by removing the D-alanine supplmentation. This is
normally done in E. coli with DAP, but D-alanine is a better target, as it is
cheaper and autoclave-stable.
1. Plan
There are 3 operons needed for the IncP system: kor, trb, and tra.
1. kor: encodes conjugative regulation
2. trb: encodes pilus
3. tra: encodes transfer proteins
In addition, we want to get rid of 2 genes: dadX and alr
1. dadX: D-alanine aminotransferase
2. alr: alanine racemase
By inserting kor, trb, and tra, while deleting dadX and alr, we create a strain
which has the ability to transfer DNA through conjugation while also having a
convenient negative selection and biosafety control: cells die without
D-alanine. We will do this by replacing dadX and alr with the desired operons.
There are 4 steps to this:
1. Replace dadX with tdk/kanR and kor. Select with kan (for kanR)
2. Replace tdk/kanR with tra. Select with zidovudine (against tdk)
3. Replace alr with tdk/kanR. Select with kan (for kanR)
4. Replace tdk/kanR with trb. Select with zidovudine (against tdk)
Step 1: kor
Run 3 PCRs with the following primers:
dadX-1-fwd AACAAGGTCTCAacaaaaaatctgcacccactttaag
dadX-1-rev TTTAGGTCTCAcgttctatcccctggcacatactag
dadX-2-fwd TTATGGTCTCTTGTCcacgtgttccttttacaattcaaac
dadX-2-rev TTTAGGTCTCAttgtaagttgcagtacccgatgc
kor-fwd TTATGGTCTCTGCTGggttttttagcggctgaaggg
kor-rev AAAGGTCTCAgacacataagcggcaagagacgaaag
dadX-1 and dadX-2 require A. baylyi gDNA. kor-fwd and kor-rev require E. coli
S17 gDNA. Assemble with pBTK622 with BsaI in a 4 part GoldenGate reaction and
transform A. baylyi. Select on kanamycin plates.
Step 2: tra
Run 5 PCRs with the following primers:
dadX-1-fwd AACAAGGTCTCAacaaaaaatctgcacccactttaag
dadX-1-rev TTTAGGTCTCAcgttctatcccctggcacatactag
kor-fwd TTATGGTCTCTGCTGggttttttagcggctgaaggg
kor-rev AAAGGTCTCAgacacataagcggcaagagacgaaag
tra1-fwd AAAGGTCTCTAACGctcgacctcgatcagggagg
tra1-rev AAAGGTCTCTGAAAccgagggcgacaagaagggc
tra2-fwd tttGGTCTCTTTTCgatgcggtgcttcttgccgt
tra2-rev AAAGGTCTCTagtaggcccacccgcgag
tra3-fwd AAAGGTCTCTtactgacgccgttggatacaccaa
tra3-rev AAAGGTCTCTCAGCagcccagctaacgcaaaaac
dadX-1 require A. baylyi gDNA. kor, tra1, tra2, and tra3 require E.coli S17
gDNA. Assemble PCRs with BsaI in a 5 part GoldenGate reaction and transform into
the A. baylyi strain prepared above. Select on zidovudine plates.
Note: tra1/tra2 junction removes the BsaI cut site, allowing for efficient
GoldenGate assembly. the tra2/tra3 junction removes the oriT site from the
genome so that gDNA is not transferred to strains.
Step 3: trb landing pad
Run 2 PCRs with the following primers:
alr1-fwd AAAGGTCTCAacaagaaacgtacctctattcatcagacc
alr1-rev TTTAGGTCTCAcgtttattcttcgtcatcatactgactg
alr2-fwd TTATGGTCTCTGCTGgcaaccgccctatacgtaaag
alr2-rev TTTAGGTCTCAttgtcagtgcctattatgagcatgc
alr1 and alr2 require A. baylyi gDNA. Assemble with pBTK622 with BsaI in a 3
part GoldenGate reaction and transform into the A. baylyi strain prepared above.
Select on kanamycin plates.
Step 4: trb
Run 4 PCRs with the following primers:
alr1-fwd AAAGGTCTCAacaagaaacgtacctctattcatcagacc
alr1-rev TTTAGGTCTCAcgtttattcttcgtcatcatactgactg
alr2-fwd TTATGGTCTCTGCTGgcaaccgccctatacgtaaag
alr2-rev TTTAGGTCTCAttgtcagtgcctattatgagcatgc
trb1-fwd AAAggtctcaaacgccatcaatcgtatcgggctac
trb1-rev AAAGGTCTCTtgctgtcgctgaattggtccag
trb2-fwd AATGGTCTCTagcaacctgtatcgcctgaccg
trb2-rev AAAggtctcacagcgcagcagcaaaaataaagcc
alr1 and alr2 require A. baylyi gDNA. trb1 and trb2 require E.coli S17 gDNA.
Assemble with BsaI in a 4 part GoldenGate reaction and transform into the A.
baylyi strain prepared above. Select on zidovudine plates.
Note: both trb1 and trb2 are rather large PCRs. They were split so that they
could be PCRed without specialty enzymes.
2. Design Outcome
The resulting strain should:
1. Uptake DNA from the environment through natural competence
2. Constitutively transfer DNA to diverse recipients via conjugation
3. Die without D-alanine supplmentation, ensuring counter-selection and biosafety
2. Progress
We are currently building this A. baylyi strain, and plan to share it openly
with any academic, commercial user, or hobbyist under the OpenMTA. We believe
that everyone should use our strains to make biotechnology more efficient.
- Keoni Gandall
Genetic Assemblies RFC 5
I-SceI Tagmentation
<Proposal>
Keoni Gandall
<keoni@geneticassemblies.com>
05 Oct 2025
<NOTE>
THIS DOES NOT WORK.
We observe the ligation rates to be approximately 55% experimentally. This can
certainly be optimized, but this reduces "complete formation" down to, at max
with native adapter, 9% of the total sample. This does not include molarity
differences.
Empirically, after one round of optimization, this simply didn't work well
enough. A single barcode instead of 3 could perhaps work much more efficiently.
You also observe this with the native barcoding vs ligation barcoding for
nanopore sequencing. Ligation barcoding creates the most reads, then Tn5 rapid,
then native, because of this ligation efficiency issue. While a single I-SceI
barcode is certainly more efficient than the native barcoding method that
requires a blunt/TA ligation (and we have empirically shown that), multiple
barcodes cannot be effectively ligated together and keep yields high enough for
robust sequencing.
However, we are looking to switch to Tn5. The reason is mostly that Tn5 can also
be used for genome sequencing, and standardizing this between both plasmids and
genomes is functionally very useful. A later RFC will have information on this.
</NOTE>
0. Introduction
We propose a plasmid sequencing method that utilizes I-SceI+ligation rather
than Tn5 for adding indexing barcodes to nanopore library preps. I-SceI is a
meganuclease restriction enzyme that cuts an 18bp recognition site. By cutting
this unique site in a plasmid and specifically ligating a barcode to its ATAA
overhang, we can produce sequencing reads that encompass the entire plasmid
rather than disparate chunks. In addition, these reads are specific only to the
plasmid, not the gDNA, so much dirtier plasmid preparations can be used without
impacting read count.
For comparison, a simple version of our method gave 70% complete plasmid reads -
or plasmid reads that represent the entire plasmid sequence without truncations
(15% were slightly shorter truncations, likely due to the physics of Nanopore,
and 5% were concatemers of the plasmid). Compare this to a normal Tn5 method,
which for a ~12kbp plasmid, gives only about 5% complete plasmid reads. Complete
plasmid reads allow reconstruction of full plasmids with far fewer reads, since
you do not need to depend on overlaps in the proper places to get sufficient
coverage.
By ligating rather than tagmenting the fragments, you also have the opportunity
to add additional barcodes. Only reads that have all barcodes will have the
final native-adapter sequence added, while the final purification reaction
washes out any barcode-only ligations. Barcodes can be lined up, (e.g.
barcode01-barcode02-barcode03), which combinatorially expands the number of
plasmids that can be concurrently sequenced without massively increasing the
cost of synthesizing barcodes.
Cheap, low-cost minipreps can also be used. For example the 3-cent per miniprep
protocol developed during the human genome project can be utilized without
worrying about genomic DNA contamination, since barcoding is specific to the
I-SceI overhang. This property can also be used to sequence low-copy plasmids,
even in the presence of large amounts of genomic DNA.
1. Previous experiments
Thus far, we have only proven this method works with a single barcode. We
resynthesized nanopore NB02 barcode with an I-SceI overhang and the
native-adapter overhang, PAGE purified with chemical 5 prime phosphorylation.
We then took control plasmids from NEB for I-SceI and attempted a
GoldenGate-style assembly reaction, where we added together the target plasmid,
I-SceI, T4 ligase, the barcode, and the native adapter to a single tube. We then
sequenced this DNA with a Nanopore flongle. This experiments is where we derive
the 70% complete plasmid number.
2. Duplex experiments
We synthesized the following oligos with PAGE purification and 5 prime
phosphorylation:
NB01_f GGTGCTGAAGAAAGTTGTCGGTGTCTTTGTGTTAACCTTAGCAAT
NB02_f GGTGCTGTCGATTCCGTTTGTAGTCGTCTGTTTAACCTTAGCAAT
NB03_f GGTGCTGGAGTCTTGTGTCCCAGTTACCAGGTTAACCTTAGCAAT
NB04_f GGTGCTGTTCGGATTCTATCGTGTTTCCCTATTAACCTTAGCAAT
NC01_f CTACGAACCTACCAACCTTATAACTCGAGA
NC02_f ACTGCAGTACTCTAGTACTCGAGTTCGAGA
NC03_f GAAGACCTCAAGACCGTAAGACCGTCGAGA
NC04_f GACTCGGCCACTCGCTTACTCGCTTCGAGA
NC05_f CCAGTCGAGGAGTCGAGCAGTCGATCGAGA
NC06_f TAAGACATGAAGACATCAAGACAGTCGAGA
NC07_f CAGCATCCCACTATCCCACGATCCTCGAGA
NC08_f TTGGTGACTTGGTCACTTGGGTACTCGAGA
ND01_f ATGGTATCATGGTAGTATGGTAGGGCTATG
ND02_f ATCCAACCAGATAACCAGAGAACCGCTATG
ND03_f TGACTTGTTCACTTGTGTACTTGTGCTATG
ND04_f GGGGTTTAGGGGTTGAGGGGTTCAGCTATG
ND05_f AGCATGACACTATGACACGATGACGCTATG
ND06_f CATGCGCAGATGCGCACATGCGCTGCTATG
ND07_f CGCGAAGACGCCAAGACCATAAGAGCTATG
ND08_f ATCGATGTATCCATGTAGAGATGTGCTATG
ND09_f TTAAGGGTTGAAGGGTTCAAGGGTGCTATG
ND10_f TTACTTGCTGACTTGCTCACTTGCGCTATG
ND11_f GCAACCACCTAACCACCGAACCACGCTATG
ND12_f TCGTCGTGTCGTCCTGTCGGTCTGGCTATG
NB01_r AAGGTTAACACAAAGACACCGACAACTTTCTTCAGCACCTCTCGA
NB02_r AAGGTTAAACAGACGACTACAAACGGAATCGACAGCACCTCTCGA
NB03_r AAGGTTAACCTGGTAACTGGGACACAAGACTCCAGCACCTCTCGA
NB04_r AAGGTTAATAGGGAAACACGATAGAATCCGAACAGCACCTCTCGA
NC01_r GTTATAAGGTTGGTAGGTTCGTAGCATAGC
NC02_r ACTCGAGTACTAGAGTACTGCAGTCATAGC
NC03_r CGGTCTTACGGTCTTGAGGTCTTCCATAGC
NC04_r AGCGAGTAAGCGAGTGGCCGAGTCCATAGC
NC05_r TCGACTGCTCGACTCCTCGACTGGCATAGC
NC06_r CTGTCTTGATGTCTTCATGTCTTACATAGC
NC07_r GGATCGTGGGATAGTGGGATGCTGCATAGC
NC08_r GTACCCAAGTGACCAAGTCACCAACATAGC
ND01_r CCTACCATACTACCATGATACCATTTAT
ND02_r GGTTCTCTGGTTATCTGGTTGGATTTAT
ND03_r ACAAGTACACAAGTGAACAAGTCATTAT
ND04_r TGAACCCCTCAACCCCTAAACCCCTTAT
ND05_r GTCATCGTGTCATAGTGTCATGCTTTAT
ND06_r AGCGCATGTGCGCATCTGCGCATGTTAT
ND07_r TCTTATGGTCTTGGCGTCTTCGCGTTAT
ND08_r ACATCTCTACATGGATACATCGATTTAT
ND09_r ACCCTTGAACCCTTCAACCCTTAATTAT
ND10_r GCAAGTGAGCAAGTCAGCAAGTAATTAT
ND11_r GTGGTTCGGTGGTTAGGTGGTTGCTTAT
ND12_r CAGACCGACAGGACGACACGACGATTAT
The ND barcodes ligate to the I-SceI exposed by cutting the plasmid.
The NC barcodes may ligate to ND overhangs and the NB barcodes may ligate to the
NC overhangs. The NB barcodes can finally ligate to the NA (native-adapter) for
loading into a nanopore device.
The structure is (plasmid)-ND-NC-NB-NA. Here is what ND01-NC01-NB01 would look like:
5' ATGGTATCATGGTAGTATGGTAGGGCTATGCTACGAACCTACCAACCTTATAACTCGAGAGGTGCTGAAGAAAGTTGTCGGTGTCTTTGTGTTAACCTTAGCAAT 3'
3' TATTTACCATAGTACCATCATACCATCCCGATACGATGCTTGGATGGTTGGAATATTGAGCTCTCCACGACTTCTTTCAACAGCCACAGAAACACAATTGGAA 5'
The left would ligate to the plasmid, while the right would ligate to the native
adapter. Ironically, there is a chance this ligation would actually be more
efficient than the typical native barcoding protocol: the Oxford Nanopore
supplied protocol requires TA cloning of a single base pair, while the shortest
overhang in our method is 4bp, with most overhangs being an efficient 6bp of
overhang.
Reactions could be done either in a single pot (like a GoldenGate reaction) or
sequentially. We designed 8 NC barcodes and 12 ND barcodes so that 1 96 well
plate (with 8 rows and 12 columns) could be barcoded. Afterwards, these
sequences can be pooled and ligated with a NB barcode, which could be once again
pooled and ligated with a single NA for loading into the nanopore device.
2.2 Failure of duplexes
3. Combination with Minipreps
This method is explicitly designed with low-cost minipreps in mind[1]. Instead
of doing alkaline lysis, this method depends on a small amount of lysozyme, plus
heat + cooling to denature the bacterial proteins / gDNA, while allowing the
plasmid DNA to remain soluble. By not doing alkaline lysis, the liquids never
dramatically change in viscosity, nor is there a large quantity of precipitation
which can be difficult to work with in an automated fashion. It does not require
filtration or beads, which makes it amenable to automation so long as a
centrifuge is provided, while also being 10x-100x cheaper. The distinct
disadvantages of this approach are solved by the I-SceI specificity.
In our own lab, we have gotten approximately 200uL of output DNA at a
concentration of 20ng/uL from an overnight 1mL of TB culture in a 96 deep well
plate, shaking at 200rpm. This is more than enough for sequencing. Importantly,
the miniprep output can also be piped into other reactions: you can directly the
resultant miniprep in another GoldenGate reaction.
We hope that this low cost miniprep is sufficiently automatable so we can close
the automation loop: we have a method for producing transformed cells, we can
miniprep and cheaply sequence those cells, and we can directly use those
minipreps in further DNA builds. This enables part reuse, which combined with
other technologies that will be discussed in future Nanala RFCs, should be very
powerful.
4. Protocol
This protocol is optimized for 96 well plates. ND barcodes are at 250fmol/uL and
NC barcodes are at 750fmol/uL in the mastermix, with the goal of having a molar
excess for approximately 20ng-50ng/uL of ~5kbp plasmid backbone from an
in-parallel automated miniprep.
Resuspend each oligo at 50uM in IDT duplex buffer. Anneal by adding equal
amounts of forward and reverse primer. For a full 96 well plate, this will
require approximately 8uL of each ND barcode and 36uL of each NC barcode. For
each column, you add 1 of the 12 ND barcodes (8 rows, so 8 wells), while for
each row you add 1 of the 8 NC barcodes (12 columns, so 12 wells). By indexing
ND+NC, you can get the row and column of any read. For ND barcodes, this would
be about 7.5uL of each to 15uL, and for ND, 24uL of each to 48uL, which is
plenty of volume to compensate for dead volume. Add 1uL of ND barcode and 3uL of
NC barcode to each well with 96uL of H2O.
This is most effectively done by placing the 8 NC barcodes in the first column
of a 96 well plate and the 12 ND barcodes in the following 12 wells. First,
aliquot 96uL of water into each well of the target 96 well plate, then add 1uL
of ND into each well across each column using a single channel pipette. This can
be done with a single tip, since the wells only have water. Second, use an 8
channel pipette to aliquot the NC barcodes into each column - noting that while
a single ND barcode will be in a column (1,2,3,4,5,6,7,8,9,10,11,12), a
different NC barcode will be in each row (A,B,C,D,E,F,G,H).
At 100nmole synthesis scale with PAGE purification and 5prime phosphate
modification we get approximately 8.24 nmoles/tube, which is 164.8uL of 50uM
stock. Assuming a healthy dead volume, approximately 12uL of ND and 50uL of NC
will be required per primer plate. This equals, approximately, 3 plates worth of
NC barcode from an average synthesis, and 12 plates worth of ND from an average
synthesis. Each individual oligo costs ~$55. 8 NC barcodes are required, so 16
oligos of synthesis, and 12 NC barcodes are required, so 24 oligos of synthesis.
In total, this is $2200 of synthesis.
In total, this costs about $300 per barcode plate in NC barcodes and $100 per
barcode plate in ND barcodes, for $400 per plate of barcodes. Assuming we can
get 80 barcoding reactions from the total of 100 barcoding reactions worth of
barcodes produced, it costs ~$5 to barcode a 96 well plate, oligos alone.
5. Enzyme cost
The required T4 ligase is approximately 0.5uL of 400U/uL for a 10uL reaction, or
200U per reaction. 100,000U costs ~$281 from NEB. For I-SceI, we want 1U per
10uL reaction (defined as amount of enzyme to cleave 1ug of DNA at 37c for 1hr),
which costs ~$341 per 2500U. In total, the cost per 10uL barcoding reaction is
approximately ~$0.70 each. Rounding up to $0.80 to account for shipping and tax,
it costs ~$80 to barcode a 96 well plate, enzymes alone.
Note: this cost can be massively decreased by producing T4 ligase or I-SceI
in-house: T4 is 80% of the total price, while I-SceI is 20%.
Further developments
We plan to release this protocol running on Nanala machines once complete.
References
1. High-throughput plasmid DNA purification for 3 cents per sample
Genetic Assemblies RFC 6
Too Much Tacit Knowledge
Keoni Gandall
<keoni@geneticassemblies.com>
21 Oct 2025
0. Background
This essay was pulled from Keoni Gandall's book "synbio25", and is published as
an RFC here as the ideas are specifically relevant and may be referenced in
further RFCs.
1. Too Much Tacit Knowledge
A major practical challenge to the centralization of physical aspects of
synthetic biology experimentation is the fact that synthetic biologists have
failed to abstract away variation between labs — in people, equipment, location,
environmental conditions — and this variance is reflected in the biology that
operates within them. This isn't a problem, it is the problem of synthetic
biology. Until we break the necessity of learning the particulars of our
execution environment, we can't (effectively) share ways of doing things. Right
now, it should be natural and expected to have a reproducibility crisis. The
requirement for tacit knowledge (implicit, unwritten knowledge) for every single
experiment extends so far that it has become an invisible requirement for
biotechnology.
In other words, it is a programming environment with zero libraries where all
compilation is done by hand, and there is no formal specification for the code,
and every single CPU works differently. To become more like modern software
engineering, one of the most scalable information-based technologies we have, we
require a few things:
1. CPUs that work (a fully-automated lab)
2. a formal specification for protocols (standardization of tacit knowledge)
3. a compiler (abstraction on top of any given fully-automated lab)
4. an environment with libraries (bottom-up adoption)
In practice, this looks like formalizing protocols as code — because code is
exactly how humans have figured out how to communicate and abstract machine
operations. Lots of people know this. But they lose the magic of what made it
work in the world of computers. The magic of lower-level languages in code is
that you can make something, quickly, that just works, and then build on top of
it. The magic of higher-level languages in code is that you can take a bunch of
libraries written by a variety of different people, then run it on almost any
machine, and it just works! Magical.
That means you can't just have protocols as code that work on a single machine
or architecture or lab. You can't just have protocols as code in a way where
people don't actually share code to build on each other. You lose the magic.
It's not just about having an API to your fully automated lab. It's about
creating the experience around those APIs that just lets people do awesome
things easily.
The magic is writing a protocol, pulling a ton of dependencies — GoldenGate
cloning, yeast transformation, plate reading — and instantly being able to
execute that protocol locally or on an automated cluster elsewhere. Simplicity
in use, sharable at the core, in a way that cannot be taken away by a single
provider. Magical!
Here is the secret to the magic that the ones who don't work at the bench do not
fully comprehend: these ideas are completely worthless without ruthless
implementation from the bottom up. You will never get there if you only
implement so-called "valuable protocols", like drug screens. They're too
idiosyncratic and one-off. No, rather you need to build implementations that
every single biologist would want to use, and can use, for their own small
projects and experiments. You need a million eyes finding the bugs to make
reliable and worthy implementations. At first, this means implementing
commodified and everyday workflows, which inherently makes for bad business and
boring papers. But some founders and investors need to make the first step, to
put up the activation energy, to produce massive value later.
As a concrete example of tacit knowledge inflating costs: right now, you can
commercially purchase clonal synthetic DNA for $125 per kbp. The price, for the
same amount of synthesis from the same company, but in the format of oligo
pools, costs $1.5 per kbp. The difference, if you do not know, is that oligo
pools are large collections of short, mutation-prone DNAs all mixed together,
while clonal synthetic DNA is long, sequence-perfect strands of a certain
sequence. The arbitrage is purely within DNA assembly and sequence validation —
which I have shown in my own lab only costs about $6 per kbp. The most
infuriating thing is that I did not believe I would hit those unit economics —
surely those big labs must have a hidden cost that I have not discovered — but I
was wrong. But it took me almost 2 years with extreme specialization and
extensive know-how in the field. And as it stands, I can't even really share my
improvements.
The real barrier to reaping these cost reductions in most biotech workflows
isn't the raw materials; it's the reliance on tacit, specialized knowledge and
the ability to keep trying to reduce costs. If we systematically handle tacit
knowledge and build environments so that protocols aren't idiosyncratic "black
boxes", we can encode this knowledge (and knowledge of all other aspects of
biological production) into code. We commodify esoteric specialization into
concrete, importable implementations. By applying this towards every single
foundational biological protocol, I hope we can exponentially decrease the cost
of doing any biology research.
This commodification will be difficult because it requires moving protocols from
human hands to robots, where the real difficulty lies not in the re-description,
but in the differences in how debugging is done. Lowering prices will require
heavy batching, which opens an economic opportunity for companies, both in
margin and in moat building.
The logical outcome of this commodification of protocols is that required
overhead for new companies is dramatically lowered, increasing the variation in
interesting companies we can see. In the long term, these factors will shift
biology towards being understood as much more of a black box, to the dismay of
scientists and to the delight of pragmatists.
<in-progress>
Genetic Assemblies RFC 7
Portable Biotechnology
Keoni Gandall
<keoni@geneticassemblies.com>
22 Oct 2025
0. Background
The problem of scientific reproducibility (specifically biological) can be
generalized to a problem of protocol portability. If protocols were truly
portable, then reinstantiation of the experimental workflow would be simple
enough that it should be trivial to reproduce any experiment with a 3rd party,
which in most cases would be worthwhile for scientific integrity.
However, this is currently not possible. Scientists cannot easily port protocols
across labs, and thus, we have a scientific reproducibility problem. The
fundamentals behind this problem are explored in RFC 6, "Too Much Tacit
Knowledge". Rather than higher level exploration of the problem, this RFC is
specific to implementation-level solutions to the greater problem.
1. Overview
The first step in making portable technology is to have some sort of interface
that any other system can rely on. We're interested in defining an interface for
working with remote labs that will first be implemented by Nanala, but may
eventually be implemented by other organizations.
We are specifically inspired by a quote from Akin's Laws of Spacecraft Design:
"The ability to improve a design occurs primarily at the interfaces. This is
also the prime location for screwing it up"
With a focus on interfaces, we wish for our portable lab interfaces to implement
a few key objectives:
1. Explicit material handling
2. Higher level interfaces implementing common protocols
3. Standardized inventory management interface
4. Defined robotic workstation interface
Together, these make a sort of POSIX for defining how a remote laboratory can
execute an experiment. We don't expect for these systems to be theoretically
optimal: they are designed to be practical and to have simple implementations.
Just like how the design of C impacted CPU design, we expect that the
limitations of these interfaces will be compensated by improvements in the
hardware running them.
The driving philosophy is to increase the legibility of operations happening in
lab while maintaining usefulness. As users create and run protocols, tacit
knowledge is converted to legible knowledge, which can be formalized. This
formal knowledge can then be ported between systems.
2. Explicit Material Handling
Physical materials (tubes, wells, plates, liquids) are explicitly handled. This
means there are no implicit or implied handling of materials: in order to use a
material on a robot, you cannot simply call for that material to be used. You
must first move that material from a specific place to a specific place (on the
robot), and then use it. While this seems insignificant, it has significant
implications for the robustness and predictability of the system overall.
By explicitly handling materials, we reduce the amount of hidden control flow
that the underlying system needs to implement while making it easier to
understand what the underlying system is doing. This makes it easier to
troubleshoot both the system (you can interpret exactly what should be
happening) and to troubleshoot a given protocol (you can predict exactly how a
material will be moved or processed). Optimizations can be made on the software
level for better handling, if that is desired - there is no reason you can't
create your own garbage collector or allocation system on top of the explicit
primitives.
In other words, we make material handling legible on each level, allowing humans
and systems to reason about materials, rather than making arbitrary use-specific
decisions on how each kind of material is handled.
3. Higher Level Interfaces for Common Protocols
Most biological experiments go through two distinct phases: strain preparation
and experimentation. In strain preparation, DNA, cell lines, or chemicals are
created to do experiments on. In experimentation these strains and materials are
used to get some form of data output. Many biology experiments share strain
preparation steps.
<in-progress>
Genetic Assemblies RFC 8
Minipreps
<Protocol>
Keoni Gandall
<keoni@geneticassemblies.com>
22 Oct 2025
0. Background
A miniprep is a protocol to extract plasmid DNA from a bacterial culture.
Plasmids can then be used for sequencing or transforming other organisms.
The art of scaled minipreps died as the Human Genome Project transitioned to
higher throughput kinds of sequencing which did not require culture in bacteria.
While single labs in those days could run hundreds of thousands of minipreps,
that is nearly unheard of nowadays, especially in an academic setting.
Minipreps are a fundamental part of synthetic biology and engineering biology,
and therefore it is essential we have a way to run thousands cheaply and easily.
Here, we describe our method for low cost minipreps. It requires no columns or
expensive reagents like magnetic beads, and can be performed with low cost
chemicals. We have run this on both a partially automated system using Opentrons
and entirely by hand. It reliably outputs approximately 50uL of 50ng/uL of
plasmid that looks clean on a gel. Rather than columns, it precipitates the
plasmid using isopropyl.
1. Protocol
1. Grow 1mL of bacteria overnight (24hr preferred) in 2mL 96 well deep well
plates with a V shaped bottom in terrific broth (TB).
2. Spin down at 3000g for 10min (or however long). Decant broth, and tap to get
the rest out. This pellet can be frozen at -20c or -80c.
3. Resuspend in 200uL of P1. Usually, this is done by adding 200uL to the top
and vortexting.
4. Add 200uL of P2. Invert 6-8 times to get full lysis. Leave for a couple
minutes.
5. Add 200uL of P3. Invert 6-8 times to get full neutralization. There should be
precipitation. Leave for a couple minutes.
6. Centrifuge for 20-30 minutes at 3000g. This should pellet the precipitate,
though there may be precipitate floating on the surface after.
7. Transfer 500uL of supernatant to a new plate. We use a deep well plate,
typically. Sliding the tip near the side of the spun-down deep well plate allows
you to avoid the precipitate on the top. If you stick it straight in, the gunk
will stick to the tip, and sometimes gets transferred to the clean plate. When
you pipette along the edge, the precipitate will stick to the side, rather than
to the tip.
8. Add 300uL of ice cold isopropyl. Mix, and then spin at 3000g for 30min. Note:
this is the most important spin of the protocol.
9. Decant.
10. Wash twice with 250uL of 80% ice cold ethanol. Be careful not to disturb the
pellet, so be gentle / aim off-centre near the bottom. Spin for 10-15min at
3000g and then decant.
11. After final decant, attempt to get remaining ethanol out. Either evaporate
or use a p20 tip. Don't over dry.
12. Add 50uL of H2O. Resuspend with a pipette or by vortexing several times.
Transfer to a pcr plate, quantify, and use downstream.
We typically get approximately 50ng/uL of pure DNA.
2. Automation compatibility
Both magnetic beads and columns are expensive. From first principles, the
approach of going for chemistry-first has massive cost-saving benefits: the cost
of the raw chemicals is approximately 1/100 of the cost of the commercial kits.
It is concievable with some chemical suppliers to get that to 1/1000 in
aggregate, while outputting a workable quality of DNA.
I hope that we can use this to lower the cost of one of the most fundamental
operations when building DNA, and that we will figure out a fully automated,
rather than partially automated, protocol.
- Keoni Gandall
Genetic Assemblies RFC 10
General Automation will beat Specific Automation
Or: CPUs over Circuits
<Opinion>
Keoni Gandall
<keoni@geneticassemblies.com>
13 May 2026
0. Introduction
Plasmidsaurus, a company doing nanopore sequencing, is the most laudable example
of a startup in the biotechnology space which has taken off. Their business
model worked because they replaced the existing Sanger sequencing infrastructure
of dropboxes with superior technology, and then got great at logistics.
Sequencing gets back to most customers within a couple hours of pickup using a
lab with many Opentrons liquid handlers and optimized workflows, and then the
data is delivered to customers in a convenient format.
This is a successful deep-tech company, bootstrapped, and profitable. But in our
mind, also limited, in that they've gotten very efficient at one specific niche
(sequencing services), without building the capacities to automate other
experiments — which is very difficult to integrate later. This limits their
ability to become a multi-billion dollar company, as the moat to competition is
relatively low — as shown by companies like Angstrom Innovation undercutting
Plasmidsaurus by 3x in their core business — and the fact that adaptation slows
when you need humans to run your lab.
Genetic Assemblies is fundamentally a services business. People pay us to do
something. What is our strategy to become a multi-billion dollar business?
The theory of Genetic Assemblies is that:
1. Aggregation of services in one lab lowers the human labor cost of any
given service.
2. A general lab system improves the velocity of developing or improving
any specific service.
1. Aggregation lowers labor cost
For idea #1, we can take an extremely simple example, loading SYBR Green. SYBR
Green is used to quantify DNA. Quantification of DNA is needed in a bunch of
different systems: both for DNA sequencing and for cell-free protein expression.
While there are economies of scale when purchasing, the simple fact that you
only need humans to load the automated system once with a certain chemical
doubles the efficiency compared to having to load the system twice in two
different labs doing two different services. You can then aggregate all inputs
to biological experiments, and see how this dramatically lowers the relative
human-labor time per experiment.
As you add more services, this advantage compounds. As you increase the number
of services running out of a single automated system, you decrease the relative
human labor needed to run any one of those individual services. At a certain
point, running many customer experiments will essentially require zero human
labor, other than the marginal cost of adding more of a given reagent.
2. General systems improve velocity
For idea #2, the simple idea is that if a lab system is developed for general
purpose use, it will naturally be optimized for research and development. For a
counter-example, Plasmidsaurus uses a bunch of Opentrons OT2 robots, which makes
for killer economics since they're low cost robots. But they are very difficult
to add more automation onto, which is fine: their economics work well with
techs+robots, and they just need to add more labs and better logistics.
For us, Opentrons OT2 are insufficient. We're starting with DNA cloning, where
there are incubation steps and centrifugation steps (the labor intensive nature
is why we have a market edge right now). Our process requires methods that can't
just run on a single robot, and so we must develop capability to handle that. As
a convenient consequence (and in part why we started where we did), our system
can handle autonomously loading, running, and unloading material without human
intervention. This allows us to run the lab as software, rather than as an
esoteric and idiosyncratic protocol to be run by humans.
We're betting that this software-based rather than protocol-based development of
services means we can iterate at a much faster pace than companies which work in
a singular niche, akin to how CPUs took over the world compared to circuits.
Circuits may be more efficient at certain tasks, but CPUs could be controlled by
software rather than hardware, and allowed far more rapid iteration speed. In
the end, tasks being defined by software changed everything. We are building a
CPU, not a circuit, for lab work.
- Keoni Gandall