Democratizing biotech research

The O'Reilly Radar Podcast: DJ Kleinbaum on lab automation, virtual lab services, and tackling the challenges of reproducibility.

The convergence of software and hardware, and the growing ubiquitousness of the Internet of Things is affecting industry across the board, and biotech labs are no exception. For this Radar Podcast episode, I chatted with DJ Kleinbaum, co-founder of Emerald Therapeutics, about lab automation, the launch of Emerald Cloud Laboratory, and the problem of reproducibility.

Subscribe to the O’Reilly Radar Podcast

TuneIn, iTunes, SoundCloud, RSS

Kleinbaum and his co-founder Brian Frezza started Emerald Therapeutics to research cures for persistent viral infections. They didn’t set out to spin up a second company, but their efforts to automate their own lab processes proved so fruitful, they decided to launch a virtual lab-as-a-service business, Emerald Cloud Laboratory. Kleinbaum explained:

“When Brian and I started the company right out of graduate school, we had this platform anti-viral technology, which the company is still working on, but because we were two freshly minted nobody Ph.D.s, we were not going to be able to raise the traditional $20 or $30 million that platform plays raise in the biotech space.

“We knew that we had to be much more efficient with the money we were able to raise. Brian and I both have backgrounds in computer science. So, from the beginning, we were trying to automate every experiment that our scientists ran, such that every experiment was just push a button, walk away. It was all done with process automation and robotics. That way, our scientists would be able to be much more efficient than your average bench chemist or biologist at a biotech company.

“After building that system internally for three years, we looked at it and realized that every aspect of a life sciences laboratory had been encapsulated in both hardware and software, and that that was too valuable a tool to just keep internally at Emerald for our own research efforts. Around this time last year, we decided that we wanted to offer that as a service, that other scientists, companies, and researchers could use to run their experiments as well.”

Lab automation obstacles

As with any interacting systems, communication is key — the systems must be able to talk to each other to achieve the desired results. To address this particular issue in their cloud laboratory, Kleinbaum and his team developed Symbolic Lab Language (SLL) to translate between instrument software platforms. Kleinbaum noted:

“The biggest issue with automating any process in this field is just that while the hardware is good … I don’t think the same amount of effort has been put into the software, so each instrument has its own (usually proprietary) software package and software language for setting it up.

“The real secret sauce of the Emerald Cloud Laboratory is an overarching platform that allows you to essentially translate to any different instrument software. For every instrument that we bring online, we have to meld it onto the system using whatever software it runs.”

The reproducibility problem

My colleague Mike Loukides has been talking and writing about the problem of reproducibility, and how lab automation is a step forward in addressing the issue. I asked Kleinbaum for his thoughts and about how his team is addressing the problem. He said automation removes the need for humans to interpret instructions, thus eliminating ambiguity:

“The biggest problem with reproducibility is not an issue of malfeasance; those are a tiny, tiny fraction of the cases. The biggest issue is that it’s a communication problem. If I tell you to mix this sample for 10 minutes, that means something very different to me than it does to you. English is actually a very poor way of communicating scientific protocols.

“By using automation and robotics, every instruction that you give translates — in the case of mixing, you know the voltage on a motor that is mixing a sample, and you know the robot is going to do it the same way every time. By reducing all lab instructions to robotic commands, you get around having to guess what the person meant by ‘mix’ or ‘spin.'”

Kleinbaum and his team are looking to address the reproducibility problem further by offering reproducibility services. It’s a two-pronged approach, Kleinbaum explained. First, they’re creating an Emerald reproducibility index, which will report data on the number of times an experiment has been repeated to offer some sense if its reproducibility. “The other piece of this,” he said, “is that we view every aspect of the lab, including the protocol for running the experiment, as just another piece of data — if there’s a public experiment that a researcher has run, you can go look at it and see every variable that they picked and can say, ‘Okay, let’s keep every variable the same except temperature and see what happens when I run it 10 degrees warmer.’ It allows you to not only reproduce the exact same thing, but to test the tolerance limits of any publicly accessible experiment.”

You can listen to the podcast in the player embedded above or download it through TuneIn, SoundCloud, or iTunes.

This podcast is part of our investigation into synthetic biology and bioengineering. For more on these topics, download the free edition of BioCoder Winter 2015.

tags: , , , , ,