Just add water...
by Tim Gardner, Founder and CEO, Riffyn
“Really Paul, just double the starting volume of the fermentation broth and you’ll get 10% more productivity from the reactor,” I implored.
Paul, the head of fermentation process development, looked at me quizzically – his eyes belaying a degree of humored skepticism. It was like he wanted to say, “here you go again, entertaining me with another harebrained idea cooked up from god knows where.” But Paul, being a hopelessly open-minded person, actually said: “Sure, why not. I’ll give you two fermenters in the next batch. If it doesn’t work, it’s just one more thing to laugh at you about.”
OK, he didn’t say that last sentence. He’s too nice for that.
A few weeks later, Paul pulled up data from the runs. “Well, it looks like it worked. How about that.” Some months later the company was running 200,000L reactors with double the starting broth volume and producing 5-10% more product over a two-week run.
This was an operational gain worth hundreds of thousands of dollars per batch – just by adding water.
I’ve had the fortune to enjoy a rare few eureka moments in science, and this was one of the most memorable. It would never make headlines, but it’s a poster child of the value of data as an asset, the impact of tried-and-true statistical methodologies, and the power of first-principle modeling.
There was no fancy AI or unfathomable genius here. This was the application of age-old methods of designed experiments, regression analysis and differential equation modeling. And it was made possible by collaborative colleagues willing to test the unexpected, access to an aggregated data set of unusual quality and accessibility, and a few good hours with JMP.
An operational gain worth hundreds of thousands of dollars per batch – just by adding water.
The model underlying this prediction was not complex; there were less than 10 equations describing the yeast fermentations. The magic was its heart – an empirical relationship between strain metabolism and reactor oxygen transfer rate that was derived from linear regression applied to 500 high-quality historical fermentation experiments.
Some weeks earlier, a talented fermentation scientist, Jake, had presented a few curiously behaving fermentation runs. He noted an odd relationship between oxygen uptake and strain performance. This led to a hypothesis, which we quickly validated against those 500 historical experiments. We then performed some multifactor designed experiments to confirm the relationship.
I felt like I had found the Rosetta stone. I could run two-week long experiments in two seconds. I could scan parameter spaces. I could test my hypotheses in no time flat.
Next, I used JMP to quickly build a statistical model of the metabolic versus oxygen coupling from these DoE studies. Then I fed that into the yeast-reactor simulation we had previously built. It was the missing link. Adding that one relationship to the model delivered near-perfect simulation of strain performance.
With fresh confidence, I began to play around with the simulated reactor conditions. It was exciting in the way only a modeling nerd can appreciate. I felt like I had found the Rosetta stone. I could run two-week long experiments in two seconds. I could scan parameter spaces. I could test my hypotheses in no time flat. I went to bed that night feeling like I could literally see the heartbeat of a yeast cell. As if yeast were me, and I were yeast.
At some point, I got the idea to play around with the initial reactor volume. I can’t recall whether it was an accident or intentional. But the outcome was compelling. Just add water, with no other changes, and you get more product out in the same amount of time. In retrospect, the reason was obvious. But sometimes we need data and statistics and math to clear the noise and see the heart of what matters.
And that’s how data, DoE, differential equation modeling, a collaborative team, and a little bit of JMP changed my life forever. Now I lead Riffyn, a company dedicated to making such beautiful, high-quality, aggregated data accessible to all.
Let's stay connected! Opt-in.
You may contact me by email regarding news, events and offers from JMP. I understand I can withdraw my consent at any time.
Dive deeper into this topic
- How do successful chemical and biotech companies implement data analytics?Visit the Time to Innovate Hub to find best practices and resources.
- Optimizing Processes with Design of ExperimentsWhat is design of experiments (DOE), and how does it deliver value? Through a case study approach, this white paper answers these questions and more.