How I Learned to Stop Pipetting and Love the Robots
My job as a biologist in the lab used to consist of moving liquids back and forth by hand. At Zymergen, I’m training robots to automate the worst parts of that job, and I’m here to tell you why you should look forward to fully automated luxury science.
I’ve been working as a genetic engineer for about a decade. It’s not as glamorous as you might think.
No one is breeding velociraptors for a theme park, or making a fleet of android replicants. In fact, it’s pretty hard to make even minor changes to the DNA of the microbes I engineer. There’s a lot of trial and error, and agonizing over experimental designs. Once I’m in the lab, most of my work involves moving very tiny amounts of colorless liquid from one place to another, mixing them to react, and then analyzing the colorless liquids they produce. If done manually, this process is laborious and slow.
Doing science with a micropipette can get tedious.
My lab work became faster with some practice and clever tricks. Before I joined Zymergen, I could do about 24 DNA preparation (miniprep) and replication (PCR) experiments in a day using a standard pipette; I could do up to 48 experiments daily if I used a multichannel pipette and swapped my test tubes for multiwell plates.
The increased throughput from the multichannel pipette was exciting, but it also introduced new challenges. If I only had a few test tubes, it was easy to remember what was in those tubes and what I had done to each. When I was processing dozens of samples in a single plate, forgetting if I was on the 36th or 37th transfer could ruin the entire experiment. Despite my optimizations, doing science by hand involved a lot of tedious human labor, was prone to errors, and was too slow for me to test all my hypotheses.
These are the (liquid-handling) droids you’re looking for
When I started at Zymergen, I had access to liquid-handling robots for the first time. These robots have automated pipette arms that can transfer a reagent into every well of a microtiter plate at once, and simultaneously plan out complex mappings between dozens of plates. I was soon performing experiments at nearly 10x the scale of my previous work.
At this new scale, I could start doing really exciting biology, gaining deep insights into the genetic workings of an organism. At Zymergen, our work relies on designing, building, and testing thousands of targeted genetic changes to optimize microbes. To find the few changes that matter amongst millions of base pairs of DNA, we have to execute an enormous number of experiments.
These experiments are only possible with automation — specifically, automated liquid-handling systems. Some liquid handlers can transfer a reagent into all wells of a microtiter plate at once, ensuring that every sample has an equal time to react. Others allow for randomizing or consolidating samples through one-well-at-a-time transfers that would otherwise be very hard for a human to keep track of.
Automated liquid handling gives us higher throughput and improved precision.
However, the tools for controlling these liquid handlers aren’t designed for the kinds of bespoke experiments that Zymergen does. We have many ongoing strain improvement programs, simultaneously engineering more than a dozen diverse microbes. These all require their own protocols, each demanding different pipette tips, microtiter plate types, and custom hardware. This is the definition of a “high-mix” production line. In Zymergen’s first few years, we amassed hundreds of protocols as we tried to meet user needs and compensate for small differences between our robots.
Ready the fleet; all hands on deck
As our fleet of liquid handlers grew, keeping all of those protocols in sync was incredibly difficult, because we couldn’t find control software designed for such “high-mix” production. We had to build our own software to parameterize the protocols and connect to our existing sample tracking system. This meant we could use two differently-configured robots (or even two entirely different robots) to run the same scientific protocols without needing to maintain programs for each. Researchers also no longer had to know the inner workings of our liquid handler control software to change simple settings like the number of plates being processed per batch.
Mind you, this software doesn’t mean that robots now just do what I want. Though I appreciate not having to move liquids by hand, I’ve had to learn how to triple-check that the alignment of twenty-odd plates piled on the robot’s deck are accurate down to the millimeter.
Loading plates onto the deck of a liquid handling robot is an annoyingly fiddly process.
Even with our sample tracking software and barcode scanners, it’s not always easy to tell if I’ve laid out the deck correctly. Sometimes it becomes obvious that I’ve made a mistake because the robot hardware crashes, scattering my precious samples. Other times, though, the program will execute correctly, and I only find out later that a swapped plate caused incongruencies in my data. This is frustrating; my experiments should not test how many times I can align a plate on the deck of a liquid handler, but rather how I’ve engineered my strain.
Towards fully automated luxury science
Which brings me to how I’m now spending my time. A few months ago, I left the biology R&D team at Zymergen to join our team of automation engineers… as a biologist. I made the jump so I could lead the biological testing of our Reconfigurable Automation Cart (RAC) systems. I see RACs as another step-change in the way researchers can think about science and experiments, akin to moving from a multichannel pipette to a liquid-handling robot.
Researchers shouldn’t have to care about what tool is being used, as long as the experiment is being performed correctly and consistently. Building software that parameterized our protocols helped us a lot, but I still had to spend too much time aligning plates on instrument decks and walking between automated systems. The RACs improve this process by always staging decks correctly and performing actions in the correct order. They don’t ‘lose count’ or ‘forget their place’ in an experiment, and they don’t get tired after working 24 hours a day.
Robot arms on a Reconfigurable Automation Cart move plates into a thermal cycler without a researcher having to intervene.
I’ve seen how the RACs can change the day-to-day experience of being a researcher in the lab. At any time, a batch of plates can be loaded for processing and the system will interleave them into its currently-scheduled work. A researcher can then walk away, monitor the system remotely, and get a notification when their plates are all processed.
My days in the lab aren’t over — researchers like me still design experiments, develop protocols for our instruments, and validate results. But the RACs are more adaptable than our stand-alone automation, which means I can construct much more complex high-throughput protocols with much less effort. RACs are currently executing portions of our strain engineering workflow, and I’m continuing to work on the robot wrangling needed to bring them to other parts of our production line.
It’s been exciting to see how rapidly the system hardware can be reconfigured when I develop a new scientific process, and how easily automation engineers can scale the number of instruments when that process hits a bottleneck. Many of the engineers on the team have been working in laboratory automation for their entire careers, and share my belief that RACs are a scientific step-change that they couldn’t be working on anywhere else.
The future I want for genetic engineers isn’t one of theme park velociraptors. It is one in which we can all go from an idea to a newly-engineered microbe without the tedium and errors of moving liquids by hand, and without being frustrated by inflexible, albeit faster, standalone liquid-handling robots. I want us to be freed to test more hypotheses and run more ambitious experiments because we’re unencumbered by trivial distractions. At Zymergen, where I’ve been able to automate away many of the worst parts of my job, it feels like that future is almost here.
Pete Kelly, a genetic engineer and Automation Scientist on the Automation Applications team at Zymergen, co-wrote this post with Tessa Alexanian, a Software Engineer on the Automation Platform team at Zymergen. Learn more about How We Do It here.