Listen on
Overview
Biotech manufacturing often struggles with fragmented data trapped in legacy systems, primarily Excel. This reliance on outdated tools prevents a unified view of production, directly impacting yields and delaying crucial insights. Yet, the path to a 40% increase in product yield through modern data practices is achievable in a matter of weeks, not years.
In this episode, host Ross Katz speaks with Yaron David, CTO and co-founder of BioRaptor. With a background spanning software engineering and an MD-PhD, Yaron brings a unique perspective on bridging advanced technical solutions with specific bioprocessing needs. He explains how data leaders can move beyond manual, siloed data to achieve competitive manufacturing.
The conversation explores the limitations of Excel in bioprocessing, the necessity of data harmonization, the role of “virtual sensors” in enhancing real-time visibility, and practical applications of operational AI to transform daily workflows. Yaron also addresses strategies for overcoming change management hurdles to achieve significant operational efficiencies.
Key Takeaways
Excel limits bioprocessing optimization and significantly reduces yields.
Many bioprocessing labs still depend on complex, multi-sheet Excel files for data analysis. This approach quickly fails to provide a unified view across batches or campaigns, preventing engineers from identifying and acting on optimization opportunities. One BioRaptor customer, after implementing proper tooling, increased product yields by approximately 40% because they could finally ask and answer questions across historical runs.
Virtual sensors provide real-time process visibility without costly hardware investments.
Monitoring bioreactors often requires expensive, specialized hardware or intensive manual sampling. Virtual (or soft) sensors offer a cost-effective alternative by training models on existing metabolic sensor data—such as dissolved oxygen, airflow, and O2/N2 levels—to estimate hard-to-measure properties like glucose levels or cell density in real-time. This increases process visibility, allowing for mid-run adjustments and faster hypothesis testing.
A full-stack data platform drives rapid ROI in bioprocessing.
Transitioning from disparate tools and manual data entry to a single, integrated platform for data ingestion, visualization, and modeling dramatically shortens time-to-value. BioRaptor boasts customer onboarding measured in weeks, not months or years, with historical data ingestion and calculation harmonization often completing within days. This rapid deployment provides immediate operational insights and enables quick process improvements.
Operational AI delivers immediate value, while optimization AI builds over time.
AI applications in bioprocessing divide into two categories. Operational AI focuses on enhancing daily workflows—automating report generation, simplifying calculations, and acting as a “bioprocessing copilot” to provide immediate utility. Optimization AI, which aims to predict and refine processes, requires a foundation of harmonized, high-quality historical data to deliver accurate, impactful predictions over a longer timeframe.
Related: CorrDyn helps biotech and life sciences companies overcome data challenges. We provide data engineering expertise to build reliable platforms and implement data quality measures. See how biotech manufacturers realize data value for more insights.
Full Transcript
Jason: Welcome to Data in Biotech, a podcast from CorrDyn where we explore how companies leverage data to drive innovation in life sciences. Every two weeks, we sit down with an expert from the world of biotechnology to understand how they are using data science to solve technical challenges, streamline operations, and further innovation in their business. In this episode, we speak with Yaron David, CTO and co-founder of BioRaptor, to explore how biotech companies can gain a holistic view of bioprocessing through real-time data integration, visualization, and modeling. In this episode, you can expect to learn why Excel is holding back innovation in bioprocessing labs, how BioRaptor enables holistic bioprocess understanding across experiments, the role of virtual sensors in enhancing real-time data visibility, and the difference between operational and optimization AI in biotech. Here we go.
Ross Katz: Yaron David, welcome to the Data in Biotech podcast.
Yaron David: Thank you very much for inviting me, Ross.
Ross Katz: Well, just to kick us off, could you give us an introduction to your background and what brought you here today?
Yaron David: My background starts as a software engineer. I was a software engineer for roughly eight years. Then I decided to do something a little bit different, went to study medicine. Probably also because of my background as a software engineer, I enrolled into the MD-PhD, so I also completed a PhD in neuroscience. When I finished the MD-PhD, I decided I do not want to be a practicing physician and that’s when my entrepreneurship stage began. Then a series of startups with a common thread of first doing good to the world, and second taking hard-to-decipher data and making it legible for the end user. That’s also what we’re doing at BioRaptor.
Ross Katz: Awesome. Can you let us know what your role is at BioRaptor and what BioRaptor does, and set the stage for the rest of the conversation?
Yaron David: Sure thing. I’m the co-founder and CTO of BioRaptor. In terms of what it does, actually, I listened to quite a few podcasts of yours, Ross. There is a common company profile nowadays, which we fit. You probably hear it quite a lot, Ross. We’re trying to make the biotech world more data-centric and help scientists make better use of their data. You can do it in many different places in biotech, and the niche that we chose to start with is bioprocess. What we’re doing is we’re really trying our best to make the bioprocess engineer’s day-to-day life better through better use of data, better data ingestion, better visualization, and data science.
Ross Katz: Just to set the stage for the rest of the conversation, can you give everyone an overview of what bioprocessing is, how it’s used in different types of biotech companies, and then maybe a little bit about the role of the biotech engineer and what they’re trying to accomplish with regard to bioprocess.
Yaron David: Bioprocess means manufacturing things by utilizing biological systems. This is a very broad term. In bioprocess the world divides into growing stuff. You grow your cells or microbes. Any type of organism can be grown. Then you harvest it. The growing it is called upstream, and harvesting it called downstream processing. Because when you grow stuff, you grow quite a lot of things. You grow what you intend to grow, but also things that you don’t need in your end product. Then you’re on your way to purify it and turn it into the finalized product that you actually want. Bioprocess engineers are in charge of the various domains of growing their organisms, making them produce what they want, and then purifying it and turning it into the final product. And making sure that the quality attributes are in place, that all the critical quality attributes are accounted for and the process runs according to the set points.
Ross Katz: Awesome. Why did you decide to start BioRaptor focused on the bioprocess space?
Yaron David: Again, I listen to quite a lot of biotech podcasts, and a lot of the times people tell the story as if they had a dream or they had a need right from the start. For us it was a little bit different. We knew we wanted to do something in biotech. We’re not biotech people per se. We’re a little bit of outsiders, software engineers coming to do something in biotech. My co-founder Ori still has it on his whiteboard in his home office. We created this board with a list of every type of process that we can think of, and we discuss with people in biotech. We had there PCR, sequencing, ELISA, gels, and somewhere there was also bioprocess. We started having more and more conversations with people, asking them what do you do with your data, how do you visualize it, how much of your data do you think you’re actually turning into useful information. When we spoke with people in bioprocess, we heard the pain. It was very easy to notice that they’re having a very hard time really listening to what their experiments, what their day-to-day data is trying to tell them. We started with bioprocess and one more thing that we realized is that there’s a very clear connection between what happens in bioprocess and the company’s bottom line. These people are actually manufacturing the end product and the better you do it, usually the better your company’s bottom line is going to be. That’s a very important thing to bear in mind when creating a startup, that you can really benefit your customers as easily as possible.
Ross Katz: That makes lot of sense. The ability to connect to the value of the improvements that you’re making to the bottom line of the company then makes it easier to sell software and optimization and data-driven processes into the industry. What did you see in your interviews, or what have you seen since launching BioRaptor that is broken about the way that biological and bioprocessing data infrastructure is done in the absence of a team like BioRaptor?
Yaron David: Wow. That’s the big elephant in the room. That’s the question that we think people should be asking themselves all the time. Some do nowadays. Most of bioprocess analysis is done by using this piece of software that was created by this company from Seattle. I don’t know if everybody knows it, Microsoft Excel. That’s really the most prevalent bioprocess software. It’s not JMP, it’s not SAS, it’s not R, nothing of that sort, it’s just plain old Excel. Excel is good when you’re starting out, when you want to analyze one run, maybe two runs, maybe 10 runs, but it very, very quickly fails when you want to get a holistic picture of a holistic understanding of what’s going on in your processes. That’s when things break down with most of the tooling. We hear it and we see it over and over again, those huge Excels, those batch records which just span 12 different sheets in Excel with interconnected formulas where you have just maybe one engineer who can roughly, roughly understand what’s going on. When you need to change anything, then all hell breaks loose. This leads to a very common issue that we see, and it is that people have sometimes stopped asking questions. They’re looking at their experiments, they’re getting an understanding of what happened during the last campaign, but they are not asking the big question about how can I really optimize my process. What happens when I’m not doing a one factor at a time experimentation? What happens when I’m trying to optimize my parameter by changing multiple levers? What if I pull on this and I push on that one? This is something that a lot of people are not doing enough of. We think that this is also part of the problem, that Excel is part of that problem.
Ross Katz: Totally makes sense. Agree that over the course of this podcast we’ve encountered a lot of situations where the data formats that are being produced by the manufacturing or assay equipment at different biotech organizations come out in Excel format, and so it’s difficult to tie the metadata between these different spreadsheets together. But one of the things that you mentioned that’s really interesting to me is that it’s difficult in the context of all of these batch processes with 12 different sheets with interconnected formulas, it’s hard to have a holistic understanding of your process. I’m interested in hearing from you, what does it mean to have a holistic understanding of a bioprocess?
Yaron David: I can give an example from one of our customers, who have been manufacturing a certain compound. I won’t name names. This is a high-value compound. They’re growing it in bioreactors and they have quite a lot of them in their manufacturing site. What they used to have is an Excel where this unlucky engineer had to write every couple of hours what is happening. They were really trying their best to gain what I call a holistic view of their world. By holistic view, again to your question, I mean an understanding of what’s happening today, as an example, versus what was happening yesterday in terms of growth. How much have our biomass been growing in the last 24 hours in a select list of bioreactors, as an example? Because these organisms who are creating their compound and they wanted to understand if they’re on track to creating this compound, so they know how to prepare for the downstream, as an example, or even understand what the supply chain is going to look like for their customers. They had this huge Excel, again Excel, and they had to sift through it and create graphs over graphs. This Excel has grown to be so large than every third time that you clicked on it, the computer just exploded. It was starting to turn on the fans and almost exploding. This was a horrific example of how people have tried to understand what’s going on, not just in the last batch, but also in the last 10 batches. But it’s very hard when you don’t have the proper tooling for that. That’s just one example of a process where once you put in the right tools in place, where you can ask the right questions and you can clearly sift through the data and ask a question, just properly ask a question: What had been the growth rate in the bioreactors which I seeded 10 days ago, and get a dashboard for that? That made a difference for them. That is something that really allowed them to increase their yield, the latest numbers are at around 40%.
Ross Katz: There needs to be a new toolkit in place in order to enable the analysis and the utilization of batches that accumulate over a longer period of time, getting that holistic view of your entire production process. Can you give us an overview of how BioRaptor supports that entire process?
Yaron David: First of all, and that’s the base layer that is needed, is to get all of the data in one place and harmonized. If the data does not speak the same language and different scales or even different bioreactors that you’re producing in are speaking a different language, then it becomes very hard to analyze the data holistically. That’s the first layer. That’s the base layer that you have to have in order to be able to then model or visualize the data. The next steps are to ask the data questions. Again, this is something that people have not been doing enough in our opinion. Just ask the data holistically across batches what had been happening. Which conditions have produced the most growth over the first half of the batch, as an example? What media parameters produced the best yield? These are just basic questions, in business it would be business intelligence. Just being able to ask those questions and get the result, and not have to wait two weeks for these results to come, that’s already something that really propels companies forward very, very fast. The next layer that you can add on top of that is be able to model the data and start understanding relationships between different parameters. You can ask broader questions, such as for all the media that I’ve used, what had been the most meaningful impact on my titer? This means that you’re not only learning from your best experiment, from your best process, you’re also taking into account where you didn’t do very well. Because what you can do in those terms is start creating a gradient, understanding that maybe less TGF-beta or less insulin produces a lower titer, but maybe some combination of these parameters would yield an increased titer.
Ross Katz: You’ve got the data harmonization that you’re doing up front, and then you’ve got the ability to ask one-off questions, the business intelligence, the observational data type approach, and then you’ve got the modeling approach. What is the value that BioRaptor is providing as you’re embedding with your customers? Is it the whole stack from top to bottom or why don’t you just give us an introduction to what a new client relationship looks like and how you start?
Yaron David: Yes. We are a full-stack company. We think that by really leveraging and encompassing the whole stack from data ingestion through visualization through modeling, is something that we can provide a lot of value to our customers. We’re a one-stop shop that allows them to do everything within the same software platform. We feel that this is part of our secret sauce. This is what enables us to provide an Apple or an iPhone-like experience because we own everything from ingesting the data from wherever it comes, having the technical people analyze it within the platform and gain value from it, and then visualizing it for the directors and executives. This makes sure that the data is very, very tight and the data has the best quality that it can have so that you don’t end up with rubbish data six months afterwards when you’re trying to model it for a better understanding. Our stack starts with ingesting the data from the devices and from the batch records, providing a visualization layer, and then also providing the data science tools that the data scientist would use and also the bioprocess engineers.
Ross Katz: Given that there is a certain amount of diversity in the equipment that different customers might use, how long does it take you to onboard new customers and get their data to a point where you can apply the visualization layer and the data science tools?
Yaron David: This is something that we really pride ourselves on. The way we do it allows us to really provide very, very short timelines to our customers. Our outlier up until now had been six days. This is six days from starting working together until the champion sends an email to our customer success engineer telling him, this had been an amazing experience. It’s something that we measure in weeks and not the typical month or years where the whole biotech had gotten used to. This is part of the ROI that you can get from these types of software platforms where it’s not something that’s going to be useful for you in two years’ time, so you should invest resources into it now. It’s something that’s going to be useful for you in a couple of weeks, and the ROI, the investment on your money and time is going to be worth it in a matter of weeks.
Ross Katz: When it comes to onboarding a customer in a matter of weeks, what does the timeline look like? How do those first few weeks look in terms of working with a customer if they’re trying to onboard you? As you put it, most onboarding timelines are much longer. I’d love a little bit of insight into what happens and on what timeline over the course of those first few weeks.
Yaron David: The first, let’s call them two weeks, sometimes less, are for the nomenclature and data infrastructure, so that our platform will speak the same language as our customer speak. This is usually achieved by looking at their legacy systems. That could be Excel. What they do is they send us their Excels and we can then create the same infrastructure within our platform. We have a lot of internal tools that we’ve built in order to make this very, very quick. We have our import template wizard where we can just configure, we feed it the Excel and then configure the field harmonization according to these Excels. This is something that can take sometimes even one day. In the third week is when we start the validation, we start ingesting historical data onto the system and the champion usually starts looking at it and saying, this is good, this still requires some tweaking. Calculations are entered into the system, so a big part of the onboarding process is making sure that all of the historical calculation that had been made, and by calculation I mean the GCR, the glucose consumption rate as an example, the OURs, which are very common in bioprocess, so everything becomes part of the platform. We have a lot of predefined calculations already in the system, so that’s also very helpful. Then the training session, training starts with the champion and then training for the whole team. Then we start doing back-to-back work where the customer would work with their usual system and also enter data into BioRaptor, and then transitioning into only using BioRaptor usually two runs, let’s call it after the first run. That’s when onboarding a team. We also have usually two months after that first onboarding, we have the advanced training session where we teach the team how to use the more sophisticated tools once there’s more data in the system.
Ross Katz: That makes a lot of sense. You mentioned the iPhone experience for bioprocessing engineers. What does that kind of experience look like in the context of the bioprocessing engineer role?
Yaron David: First I’ll say what it doesn’t look like, what the situation is like today. Nowadays people use different tools. You enter the data on one system, it gives you usually maybe zero, maybe negative value at least for the person who’s entering the data. Then there are different systems for visualizing the data and analyzing that data. By creating a full-stack system, we can provide value for all of those people which up until now have not gotten value out of the system. Let’s take the technical people. The technical people who are sampling the bioreactors and trying to make sure that the run is on track, they usually have to enter the data into, let’s call it an ELN of some sort or an Excel, and it’s just there for eternity until the FDA comes and asks for this data. It’s not being used for too many things. With us, they gain first of all a better understanding of what they’re doing, so they can immediately see the results, that could be calculations, that could be already the graph of what’s been happening and what they can expect for the rest of the run. Not only that, one of the things that the technical people have to provide is the end-of-run report. The end-of-run report is something that they have to go and compile after the run has finished and they have to take data from the media properties and what happened in the various sensors and make sure that everything was actually intact. With BioRaptor, they get it immediately without any effort because the data had been entered into the system, the report had been templated. This is generated automatically for them. That’s what we call the iPhone experience, something that you can provide when you own the entire stack of the hardware and software. It continues to the director level where people want to understand what had happened, they have to rely on good data. By knowing that the person who had entered the data has already used it before because they generated end-of-run report for themselves, they can trust the data better and they can rely on it when they’re doing their further analysis.
Ross Katz: One of the things that you mentioned in our briefing call that we haven’t had a chance to touch on yet was this idea of virtual sensors or soft sensors that replace manual sampling. As I understand it this is one of the value propositions that BioRaptor offers as you’re utilizing the platform. Can you talk a little bit about what these virtual sensors are and then maybe how they’re trained and validated by your customers as they’re being implemented?
Yaron David: This already touches on the pinnacle of the pyramid. Up until now we’ve been discussing data ingestion, visualization, maybe reporting, and now this is the modeling part. Virtual sensors allow our customers to get estimates for hard-to-measure properties during their run. An example could be glucose levels. Glucose levels or cell densities, biomass, sometimes you can’t access the bioreactor and you need a way to get this data online. One way would be to use Raman, which is a costly way, both in terms of equipment and in time invested. Raman is notorious for necessitating a lot of training runs and a lot of honing in on the model. What we can do using our virtual sensors is train a sensor which uses the existing sensors that your bioreactor already has. All bioreactors are measuring the DO, all bioreactors are measuring the airflow and are controlling the airflow, the O2, the N2, the buffer amount that is being transferred to the bioreactors. We can utilize all of these sensors to create a metabolic picture that is going to tell you the story. These virtual sensors allow you to estimate the glucose, your product level, or the biomass by utilizing all of these metabolic sensors.
Ross Katz: It makes sense. How does that virtual sensor get defined inside of the BioRaptor platform? Is this the sort of thing where these are values that are already being collected and they’ve been harmonized during the onboarding of the customer and then they just have to say I want to start predicting cell density, I want to start predicting glucose levels? Or how does that interaction pattern work? And I’m also interested in how these virtual sensors improve over time and how you estimate uncertainty around them in order to communicate expectations to your customers.
Yaron David: Yes, we start off with the currently available measurements. If the customer had already been measuring cell density and we have the sensor data, then that’s the start of it. We can use this existing historical data to start training the virtual sensor. Like everything, you divide your historical data into the training, testing, and validation cohorts. This is the beginning of it and this what brings you the first model. Those model by the way can be very, very simple. Sometimes just plain old linear regression is good enough to create a virtual sensor at certain time points. If you can allow yourself to be lenient in terms of when do you want to create a virtual sensor, you can really easily create them even without BioRaptor. That’s where things begin and you’ve asked how do we make sure that these sensors stay online without any drift. This is something that we usually tell our customers to check every time they think that there’s any major change to their process. It changes between customers who are doing it in already a manufacturing setting versus customers who are training these sensors in a process development setting. When you’re in process development, things shift around. You might be changing your control cascade, you might be changing your organism, you might be using another type of plasmid. We usually tell them that whenever they make a major change, go back, start measuring for a couple of runs again the cell density or your product and make sure there was no drift. The thing is that it’s usually very, very easy to do because you have the platform, you have the soft sensor just at your fingertips. It’s just as if you would be visualizing the old cell density measurement, you would now just be visualizing the estimated virtual sensor-derived cell density. Comparing the two is very, very easy to do. Also, you’ve asked about confidence intervals and how do we measure that? For some soft sensors, we can provide confidence intervals. For some of them, it’s a little bit harder for the ones where you’re utilizing neural networks, then sometimes the confidence intervals are harder to provide with new data.
Ross Katz: Is all of this configuration-based where, as a bioprocess engineer I’m clicking through and configuring the virtual sensors directly in the interface? Or do I need to think about how I divide up my data into training, testing, and validation or what modeling approach I should be using or what control variables I need to be applying or which target variable I should be selecting, all of these things that I would be doing if I were a data scientist in that situation?
Yaron David: I think that if we spoke in three months’ time, I could have told you that it’s all self-service and the bioprocess engineer is just guided through a wizard-like experience and everything is done behind the scenes for them. Nowadays, it still requires some hand-holding by our customer success engineer. This is something that the bioprocess engineers would tell us that they want, and we would be configuring it for them with them. But still, this is a process that takes, let’s call it a week or so, where we’re working with them to tell them where it’s possible or it’s just it wouldn’t work because you need a little bit more data and how to collect that data.
Ross Katz: I think that makes a lot of sense. It sounds like your team is partnering very closely I’m assuming during the back end of onboarding, during the advanced training portion, to establish the virtual sensors in place so that once the customer’s off on their own, that they can just take advantage of the sensors that have already been created. Have you encountered any change management challenges in terms of getting teams interested in and excited about getting out of their Excel spreadsheets and into a platform like BioRaptor to try to accomplish their goals using tools that are better built for purpose for what they’re trying to accomplish over the long term?
Yaron David: Of course. Change management is still a big issue and a big part of successful integration and the company transformation. The way we tackle it is again to provide value to all stakeholders. You can get people to change their ways in two different ways. You can beat them on their head with a big stick or you can give them carrots. We prefer to give them carrots. We try to give carrots all the way through from the technical people to the executives so that they all have a good incentive and a very short feedback loop to value. But that’s always the problem. We try to have a good champion in place, one that can really help the people and guide them in their transformation towards not using Excel.
Ross Katz: Are there any elements of the platform that you think are particularly unique or provide a particularly high amount of value based on your experience coming at this from both the combination of software engineering and the medical background and the wet lab experience?
Yaron David: One of the things that are very prominent in giving a lot of value to our customers is the full traceability across different units. From a software engineering standpoint, it’s really modeling the data through its whole life cycle. We start off with the cell bank, we provide them a way to document their thawing process, the flask, all the bioreactors, all the downstream elements, all the purification, so that you can really ask these questions across all of these unit operations and understand your process holistically. This is something that our customers love, which I love, it’s visually very, very pleasing to see this. It’s a huge aha moment. Even our software engineers love it because there’s lots of algorithms around it. By slicing and dicing through all of these trees and allowing our customers to gain insights across different batches and different unit operations, it requires a lot of tree traversal algorithms which our team loves. It’s really one of those things that once you have it, you can’t go back. You hear teams talking about it as the before and after. In the before, the downstream teams have just gotten this material and were told, this should be purified and turned into a powder. They have no idea where this substance came from, what happened in the bioreactor, what was the pH this was grown at, and with BioRaptor, they now understand everything holistically and they have context to what they’re actually doing and they can tell the bioprocess people, this clogged all the pores in my TFF. This doesn’t work well. This is because of the pH that you used when preparing the buffer, as an example.
Ross Katz: I can see the opportunity for capturing a lot of value in the design with the process in mind and the ability to see the holistic picture and to track it over time, both within a run and across runs, seems like a really important aspect of what you’re providing. If the bioreactor is a complete black box that you can’t look into and you can’t see what’s going on in there, then you don’t know what your next step should be. This means that you view it as a black box where you turn on on Monday and you harvest on Friday and during the whole time you can’t really know what’s going on in there. Are my cells happy? Are they not so happy? What should I be feeding them more? Should I be feeding them less? By reducing this opaqueness, by making it very, very easy to understand what’s going on in there, they were able to create better experiments so that they knew that if they’re seeing the cell density stagnating, as an example, there’s not a lot of growth, that they should be trying something different. You don’t stay the course if something is wrong. This is something that they couldn’t have done before. This really compress their research cycles. Ross Katz: I think that makes a lot of sense. Let me say it back to you to make sure that I’m understanding correctly. Prior to that, when the bioreactor was a black box, then you could create these experiments and then you would be able to maybe see the outcome at the end when the process completed on Friday, but you didn’t have the level of visibility into the entire production run that you have after you’ve implemented BioRaptor. Having that richer information about what’s going on inside of the process as you put it, the level of happiness of the cells inside of that process as you’re going, it allows you to then qualify the outcome of that experiment in a way that allows you to be more intelligent about the way that you design experiments and then get richer information back as a result of that experiment. Am I thinking about that right?
Yaron David: Yeah, yeah, yeah. I think in software engineering we have this notion of fail fast, in startups in software development. You want to get to test your hypothesis as soon as possible. In biotech, especially in big pharma, it’s not a very prevalent notion. We like to think about it as the experiment within the experiment. The experiment doesn’t start at day zero and ends at day six. There’s a lot of opportunities during a run to change things and tweak things to make sure that you’re on the right course. When you have a black box, it’s just not possible. You have to open that box in order to be able to squeeze more information out of your run.
Ross Katz: I think that makes a lot of sense. It’s the lost connection between diagnostic reporting, business intelligence, and design of experiments and experimentation that I don’t think is talked about a lot, the way that these two things feed into each other and the richness of data that you’re able to consume and how that impacts the value of the experiments that you’re running, even if you’re not able to Bayesian optimize the entire process out of existence.
Yaron David: Nicely put. Yeah, yeah, yeah, exactly.
Ross Katz: Interesting. As we head toward the end of our conversation, I just want to zoom out a little bit and ask you a few questions about the ecosystem at large. Is there a belief about bioprocessing or biotech data in general that you think is widely held but dead wrong? A misconception that you encounter regularly throughout the industry?
Yaron David: One of the things that we see a lot with companies is not looking at the raw data. They run their bioreactors, they read a lot of data, they sample every 10 seconds or so, and then they just discard this data. As if it’s only used to make sure that the run was alive and it does not contain any meaningful information in it. I think this has a lot to do with the lack of tooling in understanding this data. We are starting to see this changes with the model predictive control and tools like BioRaptor which allows you to really squeeze more information out of this data. But throughout the industry, this is something very, very prevalent, just not using any of those sensors. This is something that we think would change a lot in the next couple of years. One more example around this is that you would think now to not run a bioreactor and having a control cascade to set the dissolved oxygen at a certain setpoint. You wouldn’t think that this is something you would need to manually tweak, and this is something that’s going to be changing I think in the way we grow cells in bioreactors where it’s going to be a lot more adaptive. This will only happen once people will start understanding their sensor data better.
Ross Katz: That’s interesting and it reminds me of a situation I’ve had recently at a biotech organization. I’ve seen a similar idea over and over again where especially in biotech manufacturing, you take the values that your machine has given you, the summarized values at the end of the run as being all the information that you need to know. But if you’ve taken a sample of what’s happening inside of the machine every 10 seconds as you put it, or in the case of some of the machines I’ve been working with every 10 milliseconds, there’s a richness of information in there. But as the data volume grows and the tooling isn’t available to help you ask the right questions, help you visualize the data in the right way, help you to get the insights that you need, detect the anomalies that account for the variability that happens within that sampling regime, there’s a gap between both the tooling and the talent that’s available at the manufacturing engineer side and the types of questions that need to be asked in order to generate the insights from those machines. I’ve definitely seen that myself.
Yaron David: A data impedance mismatch maybe.
Ross Katz: I like that terminology. One of the trends in the industry that I’ve heard about is this increase in the adoption of single-use technologies in bioprocessing. I’m wondering if that impacts you or impacts BioRaptor in any new way, or is it really just about the bioreactor and the data that it throws off?
Yaron David: BioRaptor we have customers using both types of technologies, using a lot of single-use and a lot of multi-use stainless steel type of equipment. I think that with single-use because you’re paying more for those single-use bags, you really want to optimize throughput. You can’t allow yourself to be limited by data analysis and just having to sift through the data to make sure that your batches are right. If you’re optimizing for throughput, you also need to optimize for throughput in your data analysis. I think people would value their time more when they’re using those single-use bags than using stainless steel because again you have to make sure that your sprints, your campaigns are back-to-back together and there’s no dead time between those batches.
Ross Katz: As the expense of each successive run goes up, then the value of the data and the importance of optimizing that process goes up as well if I’m hearing you correctly. The last question before I let you go, are there any emerging technologies or upcoming advancements that you’re seeing on the horizon that you believe will have a big impact on the work that you’re doing at BioRaptor?
Yaron David: We’ve been speaking for 50 minutes now and we haven’t said AI. I don’t know how we managed to conduct an adult’s conversation without saying AI so many times. I think the progress that’s been made is amazing and just the term AI, it’s so evolving all the time and it means so many things to different people at many times. We’re embedding AI in many, many flows nowadays from the day-to-day, from creating the reports, from creating calculations to just assisting you as a bioprocess copilot is another use of AI and the virtual sensors. I think this is one of the main things that are evolving nowadays and people have to think how they will be doing AI not in two years, but tomorrow morning. AI is no longer a C-suite strategy, but it’s something that can be embedded in the day-to-day lives of bioprocess engineers. This is something big that’s happening now and is going to grow even bigger in the next couple of years.
Ross Katz: What I’m hearing from you, if I understand you correctly, is that the biggest opportunities for AI in the work you’re doing with BioRaptor is in smoothing the utilization of the interface for end users so that they understand how to use the platform or they’re able to operate more complex aspects of the work that you do inside of BioRaptor without needing someone from your support team to train them or implement something on their behalf. Is that correct or are there other use cases for AI that you’ve used as being really potentially valuable to you over the next few years?
Yaron David: Yes, you’re right. This is what we call the operational AI. It’s not the optimization AI. It’s not the AI that will allow you to optimize your process necessarily. This is operational AI and this is the tomorrow. It’s not the AI that you’re going to be using in three months’ time, it’s the AI that you’re using tomorrow. This is how we divide our use of AI in BioRaptor, the operational and the more long-term optimization type of AI.
Ross Katz: Because the long-term optimization type of AI needs to accumulate data over time in order to be able to make the predictions accurately. Am I thinking about that right?
Yaron David: Exactly. Yeah, yeah, yeah. Exactly. You need to have a harmonized data for that customer before you can apply these types of AI. But the operational AI, that’s immediately available.
Ross Katz: That makes a lot of sense. Yaron, it’s been great to have you on the podcast. Really appreciate the time. For people who are interested in learning more about BioRaptor, where should they start if they want to learn more?
Yaron David: Our website is great. Feel free to browse or connect with me. I’m on LinkedIn or email me at [email protected].
Ross Katz: And the website is bioraptor.ai as well. Well Yaron, thanks again for coming on. I really appreciate it and look forward to connecting down the line.
Yaron David: Thank you very much, Ross. Thanks for inviting me.
Jason: And that’s it for this episode of Data in Biotech. If you enjoyed the episode, please subscribe, rate, or leave a review in your podcast player of choice. See you next time.
