Early in our education about science, we learn that to understand what is happening in a system, we can only change one variable at a time. That seems reasonable. By doing that, we can see which variables matter.
It turns out that this is not only slow, but in biology, possibly incomplete. If there is a two-factor interaction (not uncommon) you will miss it. The good news is that it’s possible to analyze many variables at the same time and understand which ones matter. One can also see the interdependence of multiple factors.
Markus Gershater and Claes Gustafson joined me to talk about Design of Experiments and how it can be used to understand and optimize processes with multiple variables.
…particularly with biological systems, often the best setting for a particular factor will depend on the level of another factor, right?
This is called the two-factor interaction. And when you are only investigating one, each factor in the isolation of all of the others, you'll never see those. … a fundamental feature of biology is this kind of interaction and you'll just never see that doing one factor at a time. - Markus Gershater
As I think about it based on my own experience, a long time ago in a galaxy far away, a lot of academic science was focused on understanding one narrow slice of biology. (Is this gene important for pathogenesis?) Scale and automation had not yet entered the picture. I’m not sure I even imagined it.
But in this galaxy now, there are all kinds of biology being done at scale. Science can be done faster, but scale can also be more expensive. It makes sense to optimize processes, not only to produce a better product but also to reduce the cost. Design of Experiments can do both. You can use it to make a better antibody, enhance the yield of an enzyme and/or reduce the cost of the media used in either of those.
The magic of DoE is how we can get the most information from the least data.
Sorry to interrupt here, but you are subscribed aren’t you?
Interestingly, we don’t need to test every value of every variable. If you have two values for six variables, 26 is 64 but you don’t need to test 64 combinations. Sixteen may be enough to see what’s important (because math) and then you iterate to hone in on a solution from there.
In some (many?) cases, you don’t need the best solution, you need one good enough to commercialize, e.g. an antibody with a ridiculously low Kd .
…if you just take the simple example of an antibody, you wanna make sure it's, it, it certainly binds to your target at the Kd that makes it relevant. And you wanted to make sure that it can be produced at a titer that is high enough in your CHO system and you wanna make sure it has a melting temperature that is above a certain degree and you wanna make sure it doesn't aggregate at a concentration of x… I mean, who cares if it binds? What you truly care about is, does it shrink the tumor? - Claes Gustafson
While much of this may be new to biologists (who don’t like math), this is chemical engineering 101. It can be used to optimize microwave popcorn or your route to work. (Listen to the interview). My marketing friends could use it to optimize variables in an email or on a landing page.
Believe it or not, this concept is close to 90 years old. Now would be a good time for more biologists (and popcorn lovers) to take advantage of it.
It’s also an awesome time to share this podcast with your colleagues. Thanks.
Markus Gersahter is the CSO and Co-Founder of Synthace.
Claes Gustafson is CCO and Co-Founder of ATUM.
Conversations are fun and your deepest insights are your best branding. I’d love to help you share them. Chat with me about custom content for your life science brand.
Share this post