Engineering biology is an iterative process. Design. Build. Test. Learn. Rinse and repeat. It can involve a lot of pipetting, picking colonies, screening a large number of variants, and other highly manual and repetitive tasks. What if biologists could outsource these tasks to robots?

While many in the field have explored using robotics to streamline their processes, it’s been difficult to adopt. There’s a steep learning curve, instruments are expensive and automated tasks still require human intervention.

Last month, I caught up with two people trying to change this: Roya Amini-Naieni, CEO and co-founder of Trilobio, and Brian Frezza, co-CEO and co-founder of Emerald Cloud Lab. Trilobio creates modular robots designed for minimal human intervention. When I asked Amini-Naieni if she thought there would be a time when biologists wouldn’t know how to pipette, she responded without hesitation: “Yes.” Emerald Cloud Lab provides a centralized remote-controlled lab that allows scientists to perform wet lab experiments in a Cloud lab from anywhere in the world. Frezza believes that by centralizing lab automation, they can overcome the cost and training hurdles that prevent scientists from adopting automation.

Below, hear their thoughts on the current state and future of automation in synthetic biology. 

(The interview has been condensed and edited for clarity.)

Can you provide an overview of your company? What specific issues or inefficiencies are you trying to address?

Roya Amini-Naieni: We started our company based on the premise that there's a reproducibility crisis in the fields of synthetic and molecular biology. Our goal is to make it so that fully automated biology labs are normal through what we like to think is a groundbreaking robotic architecture and software stack. 

Right now, incumbent machines on the market are isolated. A liquid handler processes samples and moves small volumes of liquid around. An automated freezer spits out samples for biologists. But, the biologist has to transfer samples from the freezer to the liquid handler so a human has to be in the loop.

We've taken the biology lab and broken it up into a bunch of robotic modules and developed an IO where our modules—we call them trilobots – snap together like Legos and can instantly transfer materials between each other. That's how we eliminate the human intervention required.

Brian Frezza: We are a programmable laboratory where you can do everything via computer code to describe how you want your experiment to run. We run both automated work cells in our center and manual tools if people want to use them. You program in exactly what you want to have happen in the lab, and then we'll execute it exactly the way you want. The notion here is that anyone can run any experiment at any center and get exactly the same result. What our software is doing internally is managing the whole floor, so it's able to keep this whole lab running 24/7, 365 days a year.

The problem we're trying to solve is shortening the distance between “Here is my idea for the experiment” and “Give me my data back.” Scientists want to get more data through with less effort. And they want to do it more nimbly.

‍

How do you ensure automation is accessible and user-friendly for researchers with varying levels of technical expertise?

‍RA: We believe that biologists shouldn't need to know how to code. We are making a graphical interface for biologists to define experiments. We're trying to give biologists maximum control over what their robot does while also abstracting away the need to code entirely.

‍BF: Most of our users do not know how to code. They're scientists who know their science really well. What we had to do to get it to feel like you're kind of just using instrument software. We made it so that it's a no-code system, where as you select what you want to have happen in the graphical interface, it’s being recorded in code. This is what you can email to someone else and get it to behave exactly the same way. 

‍

What do you think are the most significant bottlenecks or issues in synbio that automation is helping to address?

RA: Doing manual labor in genetic engineering is so repetitive and so prone to error. Your entire experiment could go to waste, and you could waste a bunch of money and time. We want to eliminate the psychological burden that biologists bear as they're doing manual labor. This also stops them from being able to design new experiments because the manual labor is what biologists end up focusing on in the lab. Because of it, they stop being able to design experiments or think as creatively.

‍BF: I don't think it's unique with synbio as for any other form of laboratory sciences. The heart of what everyone is trying to get to with automation is they want better reproducibility. They also want things to be higher throughput so they can get more data through it. 

‍

What about AI? How do you see AI and automation working together, and what are the challenges and concerns?

‍RA:  AI can be used for troubleshooting experiments automatically. I think that's the most useful application because biologists run experiments, and their experiments will fail and succeed for unknown reasons. 

‍BF: AI is a great tool for allowing scientists to move up the abstraction layer. So as AI gets smarter, we can get it to do more and more of the bottom level at a higher efficiency than we do as humans as we think about more top-level ideas. 

Imagine me being able to sit here and say, “All right, let's pick a couple of things like gradient and flow rate and maybe temperature and the column.” Then you give the computer some degree of freedom to solve it. You're allowing the AI to figure out how to tune all those settings to what they need to be. That's really exciting because it means you're not actually designing an experiment. The AI is going to design the experiment. In many ways, AI and automation play the same role.

‍

What are some of the challenges in adopting automation systems in labs? What advice would you give to synbio labs looking to add automation to their synbio workflows?

RA: The reason we started Trilobio is because it's so painful to adopt automation into a lab. As an example, if you automate one experiment and you move your robot a little bit, you have to recalibrate it. You can't change that experiment, because then you have to rewrite all of the machine code, and then you spend weeks doing it. I know companies that have spent millions of dollars on fully automated, integrated systems that they don't use because of how inflexible and hard-to-use they are. So that's kind of what we were trying to address with Trilobio and eliminate that upfront cost as much as possible in adopting automation. 

BF: Everybody kind of agrees on what the upsides of automation are, but the reasonable pushback people will give you is that it's very expensive. You have to have dedicated experts in just those things to bring things online. If you do it in your own lab, you have to get everybody trained. So Cloud is an interesting thing, where you solve two of those problems outright. 

‍

What would you say are the top processes that benefit the most from automation?

RA: There are experiments that are just really prone to human error, like if you’re hit picking a bunch of samples. Experiments that are just ridiculous for a human to do are very good for automation. Then there are experiments that you want to increase throughput. If you are doing bioprospecting research and you want to search hundreds or 1000s of microbes, that’s really good for automation.

BF: Anything where scale or reproducibility matters tremendously are good use of it. Automation has natural computational closure around it. The robots can't understand ambiguity so it's going to give the same performance when you run it. There's a reason liquid handlers are excellent for handling 96-well plates. You should probably never, never pipet those things by hand.