Hard Biology: Reliability
There's a million reasons why biology is hard tech. Today we're talking about one of them: Reliability.
Transcript
Biotech R&D is complicated and sometimes things go wrong. That's not a big deal. We all know what it's like to be doing stuff, something breaks, we fix it, we move on. The occasional error doesn't make something hard.
And yet, when you talk to a bioprocess engineer about reliability, sometimes they get this thousand yard stare, like they've gazed into the abyss of hard tech and it changed them. What are they seeing that us normal people don't?
I think it comes down to the difference in reliability between a single operation and a series.
Biotech R&D is a series of operations. For example, maybe you're developing a biologic - an antibody drug. You've probably got a library of candidate antibodies, each designed with a slightly different sequence. You want to generate some lab data for this library so you can select the best one to advance toward the clinic.
So what do you have to do? You've got to assemble the DNA sequences that will encode the candidates. You have to transform that DNA into living cells. You've got to grow the cells and get them to produce the antibody. Then you've got to break the cells open, purify the antibodies inside and characterize them with a series of tests. In this case, you might care about properties like binding affinity, expression level or thermostability.
Each of these steps includes multiple operations. A mechanical movement. Liquids coming in and out. Mixing. Taking a measurement. I don't know the exact number of operations, but it's a lot. Call it a hundred, depending on how you count.
That sets up a simple mathematical argument. If you have to perform a series of operations, and any single failure means the whole process fails, the overall success rate is obtained by multiplying the success rate of individual steps. In other words, it declines exponentially as the process gets longer. Failure rates like 1% that might seem small in the context of everyday life explode and kill you when you chain them in a series.
The bioprocess engineers out there know this, of course. And they have all kinds of strategies for dealing with it. You can add redundancy to the process so that individual errors don't matter as much. You can monitor and correct errors in real time. Or you can just get really good at engineering so those failure rates get way lower than 1%.
Engineers work miracles. They really do. But miracles take time. And it turns out, you usually spend a lot more time perfecting the process than standing it up in the first place. It adds friction for anyone who wants to do biotech R&D at scale. And the upshot of this is that we, the biotech industry, are not really using automation effectively.
On the one hand, you see really big and heavy workcells dedicated to a single process. These are worth all the time to set up and debug, but only if you need a really huge dataset against a very narrow problem. You've got scale, but not flexibility. You might lock down one of these bad boys only to find that the data you need doesn't quite fit the capabilities. You want data now, not 18 months from now.
On the other hand, some people solve the process reliability problem by just not really trying. No automation. Or just a couple of benchtop robots. Human hands make plenty of mistakes, but they can also continuously monitor and correct them. So you get the job done one. sample. at. a. time. I've seen companies out there offering to characterize antibodies in tiny little sets. Like five at a time.
That's the natural scale of a human hand. But you're just not going to find the most effective therapeutics if you're doing five experiments at a time. You're just not. You've got flexibility, but not scale.
So the answer, of course, is Ginkgo Bioworks and the foundry model. You can solve a bioprocess reliability problem by partnering with somebody who's already solved it. Here in the Ginkgo foundry, we can automate an R&D workflow once and then keep it running for many different applications.
Its lab data as a service. It works because many workflows in biotech R&D are similar across projects: Assemble - Transform - Grow - Lyse - Purify - Measure. Your R&D project includes some of these operations, and a few more besides. The DNA sequences are different, maybe the cells are different and the substrates. Your therapeutic target is extremely unique.
But most of the R&D operations aren't. And so most of the bugs that make reliability engineering so challenging, we've seen them already and we've squashed them. The hard tech of reliability engineering that gives your R&D team that 1000 yard stare, we've already been through that, friend, so you don't have to.