Just when you clear one process bottleneck, it seems you’ve created three more. Advancements in computer aided biology and machine learning allow scientists to automate design, often resulting in more designs than a single researcher can test reliably by hand. Fortunately, robots are here to help.
LabGenius, a growing UK synthetic biology company, relies on automation to generate robust data for their AI protein design platform, EVA. With these tools they have created a high-throughput process for therapeutic protein development. Automation plays a vital role in forming their discovery cycle and delivering high quality data to inform and improve their protein design platform. I caught up with LabGenius Software Engineer Staffan Piledahl to learn more about their Analytik Jena workcell.
LabGenius has developed EVA, an AI-driven discovery platform that uses machine learning models to artificially enhance protein designs. These models use public and LabGenius’ own data to predict better protein designs, for example proteins resistant to proteases in the gastrointestinal tract. Where natural evolution is a tinkerer, EVA is very much the engineer. The platform is improving all the time, but it requires huge amounts of data in order to build on previous successes. LabGenius learned early on that automating the Design-Build-Test-Learn steps that underpin their discovery cycle would not only increase the amount of data generated, it would also improve the quality of that data.
Piledahl explains the importance of the quality, “We train machine learning models based on that data. EVA is able to interact with these models and create optimized designs from them. The real value is in the data.”
EVA then produces a proposed DNA library. DNA oligonucleotide production is outsourced but the library itself is assembled in-house. The DNA parts are designed with an assembly method in mind, such as Golden Gate, so they can be easily assembled by LabGenius’ robots. Automation is well suited for the build cycle as the protocols for different assembly methods are well-defined.
“We have different methods for assembly; some developed in-house and some are standard. The next step after Build doesn’t really care too much how that library came to be. So, if EVA needs to use Golden Gate assembly or USER assembly or something else, I just need to make sure we have automated protocols for these pieces.”
LabGenius has made an automation workcell to suit their needs. It is comprised of the Analytik Jena CyBio Felix, CyBio Carry, and BioMetra TRobot thermocycler. Additional hardware, like a 96-well plate reader, can be incorporated. These machines are connected through the composer software provided by Analytik Jena and third-party devices like the plate reader are seamlessly integrated into the workflow by themselves, no external support is needed. The CyBioCarry system is truly plug and play.
Automation was supposed to give me walk away time but here I am again watching the robots work @labgeni_us.
Amazing to see everything logged, variables tracked and experiments reproducible. Ironic that it brings feeling of being in control? #automation #analytikjenauk #synbio pic.twitter.com/X32ICWU7Cd
— Staffan Piledahl (@Staefy) March 19, 2019
The CyBio Felix liquid handling system has a two-level deck setup with 12 spaces available. Here, pipette tip racks, reagents, cooling blocks and plates can be set up to suit the needs of the method. Their Felix has two adaptors for the pipette head (though more options are available). The CHOICE head has 8 channels and is used for the discovery cycle while the 96/1mL and 96/250μL heads address the full 96-well plate and are used for scaling up during protein production methods.
The TRobot thermocycler is specially designed for use with automation, to allow easy access to the plate for the robotic arm of the CyBio Carry. The platform is highly adaptable. A surprisingly complicated challenge in setting up the workcell was around sealing and unsealing the 96 well plates when used with the TRobot thermocycler.
Can't fit a sealer/peeler in your work cell? Don't want to dish out the $$?
Here's how it's done at @labgeni_us
Using #comfortlid from @HamiltonCompany . Name is a bit weird tho. Sounds like something I'd need after a break up? #synbio #automation pic.twitter.com/WyH41WV9P4
— Staffan Piledahl (@Staefy) April 10, 2019
Piledahl explained their solution, “When you put a plate into the PCR machine, you usually put a seal on it. In order to do that automated, you need pretty extensive and costly equipment, such as an automated sealer and peeler. We found a way to use a film lid that is pushed down by the TRobot, creating a seal over all the wells, and this lid can be lifted off and on by the robots.”
The (un)changing role of the Biologist
While the Build step is quite straightforward, the Test step can be anything but. The CyBio systems are highly adaptable and are set up to be “plug and play.” If the testing team need to optimize their overall process manually, they can still automate the known parts of the process to save time and ensure reliability. For new processes, the automation and testing teams work closely together, with the automation staff doing some work on the bench by hand to get a better idea of the experiment’s needs. For anyone concerned that automation will detract from the role of the scientist, the truth is quite the opposite.
Piledahl shared his insight into the truly symbiotic synthetic biology relationship between the two teams: “There is always a need to first stabilize things and make it work on the bench manually. What happens is that biologists don’t have to spend their time running the platform anymore. They spend their time going after new targets, optimizing workflows, making things better. They’re now enabling the expansion of the platform.”
The results of the test cycle can then be fed back into the machine learning models, allowing them to improve their design capability and ultimately use EVA to generate an improved DNA library to test. For Piledahl, the automated workflow is less about speeding up the process and all about generating reliable, reproducible data.
“It’s way more important with data quality and reproducibility, and having variables tracked digitally, so we can correlate them with the result and the performance of the models,” he explains.
Each step is digitally time-stamped and that level of reproducibility is key. Even though it wouldn’t make sense from a throughput perspective to run a single sample with the robots, it makes sense in terms of reliability. The variation introduced by manual working seems huge when compared to the precision of the automated system. With each timestamp you know precisely how long a heating step took as well as the ambient temperature of the room. All of this is recorded in the Analytik Jena software’s logging files and can be correlated with your data at the end. Given the number of optimizations performed, this feature is by far the most important.
“It would be completely impossible to do by hand,” Piledahl says, “because if that wasn’t tracked I … [laughs] I can’t even imagine! I can’t do these experiments when there’s 15 wells with different conditions without making mistakes. Doing a whole plate, I can’t even think about.”
Automation has historically revolutionized every field it enters, and biology is no different. For LabGenius, their Analytik Jena workcell has introduced the reliability and precision needed to build on their EVA technology and cleared the bottlenecks in discovery and development of therapeutic proteins.3