Denise Pohlhaus - Computational Sensory Science

Welcome to "AigoraCast", conversations with industry experts on how new technologies are transforming sensory and consumer science!

AigoraCast is available on Apple Podcasts, Stitcher, Google Podcasts, Spotify, PodCast Republic,Pandora, and Amazon Music. Remember to subscribe, and please leave a positive review if you like what you hear!

Denise Pohlhaus leads a team of data enthusiasts within GSK Consumer Health. Denise has a PhD in Chemistry from UNC-Chapel Hill and conducted her postdoctoral studies at UCSF. She started off as a computational chemist doing pre-clinical drug discovery within GSK Pharma R&D where she applied physics-based methods and cheminformatics techniques to drug discovery problems -- resulting in patent applications and publications.

Denise then moved to GSK Consumer Health to build and execute statistical capabilities for a team of consumer innovation and sensory scientists. In her current role, she uses traditional statistical methods and advanced analytic approaches to understand the desired consumer sensory experience and to help guide product development scientists on how to formulate products that exceed consumer expectations. Denise is passionate about building processes and strategies to ensure data are clean, standardized and high quality to ensure the best predictive models can be derived from them.

LinkedIn

Transcript (Semi-automated, forgive typos!)

John: So, Denise, welcome to the show.

Denise: Thanks, John. Thanks for having me.

John: Great. Now, before the call, Denise, we were talking about something I thought was very interesting that you were describing how in the chemistry community you witnessed a kind of transition towards computational chemistry and you see something kind of analogous happening now in the sensory community. So I think it'd be really interesting for our listeners to hear you kind of start with the story of how you saw chemistry evolve and then what sort of changes you see happening in sensory that might mirror some of those evolutions.

Denise: Yeah. So just to caveat it. It sort of predates me. I'm not so old that I saw the birth of computational chemistry, right? But traditionally there had been the space of chemistry where you had organic chemists going into the lab and making crazy solutions and concoctions and things. But over time, particularly in the drug discovery space, it became clear that they needed to do more with the data that they were generating, particularly as within the chemistry space and biology as well. We were able to create more and more different kinds of data. And where you have millions of rolls of data, sets of compounds that are being made with some link to biology and so coming out of that was a sort of thinking more from the physical chemistry and physics space of how do I use this data in a really sensible way and use that to kind of make predictions on what kind of compound I'm going to make next. And so those two kind of fields kind of smashed together. And over time, I think around the 80's, really the field of computational chemistry really kind of emerged coming out of that. And what's so super interesting to me is so a lot there's many different flavors of computational chemists, but the really good ones are people who have a strong understanding of chemistry and biology, but also have a strong understanding of computational methods. And what I'm seeing in the sensory space is a similar kind of emergence where we are seeing sensory scientists who are classically trained, a sensory scientist, really learning more and more coding skills as well as statistics and kind of becoming this new hybrid role in person where they're using all the strengths of their field to really supercharge the sensory space. So that's kind of what I was thinking, and we're seeing that in our team as well, where we're bringing in folks who really are trained in more traditional sensory science, but had done some coding in school and so are now kind of utilizing their skills in kind of this new space.

John: Yeah. Now, it's really true what you're saying is so fascinating because, I mean, for one thing, there's just okay, even before you get into the additional analytic capability that you might have from using some of these computational methods, there's just an efficiency gain that I think is going to be hard for anybody to compete with. If you've got kind of reproducible research where you know what the steps are, I mean, I think we've all had this experience where you have some report and really God knows what was in the report, you know, several years later. Whereas, you know, some of these tools from data science are helping to lead to kind of reproducible outcomes. To be interesting for our listeners to hear you talk a little bit about reproducible research. And, you know, what are some of the steps that your team is taking to help make the research within GSK be actually really reproducible?

Denise: Yes. And I think this is why it's so powerful to have these individuals who have these skill sets come to kind of embrace data science for their own, because fundamentally, you cannot build a good model unless you have data that is clean, standardized, and you kind of understand how the data was generated, right? Because this is kind of part of understanding whether a data set is reproducible. So you want to understand, you know, what are the references that were used when they were defining their lexicon. How reproducible were the replicates across different sessions. And what I like about having people who have a deep knowledge of sensory science, this work is they can quickly understand where the gaps are in the data and in producing the data. And when you have them on board as part of your journey, they become then your passionate advocates for wanting data to be standardized. And I think that just kind of enriches the space and makes it so much easier to get by and to have good data standards and good data quality.

John: Right. Now, I definitely agree with that. I think we've all had this experience. I mean, there's this joke right? In the data science is 80% cleaning data and 20% complaining about cleaning data.

Denise: Yeah, which you and I have experience in our professional lives, right?

John: Yeah, definitely. Yeah, I mean, I think that's definitely true. Yeah, it's something that I've certainly had this experience and maybe you can speak to your own experience of this as well. That as you're setting up a pipeline, there's this idea that it should be the case, that you put the data in the top of the pipeline and then you turn the crank and, you know, some sort of standard analysis happens. Some reports come out. But, you know, in reality, in the wild, data are often not standardized. So what are some of the things you're thinking about as you're pushing to standardize more of the data within GSK? What are some of the processes you're putting into place or some of the procedures that you're asking people to follow to standardize the data?

Denise: Right. So I spoke about the lexicon in the beginning. That is very important, right? Because you do want to have something that's a bit more set. So I don't want to go into a long discussion about absolute scales versus QDE because I am not equipped enough to have that discussion without embarrassing myself. But certainly I do appreciate having references as part of the lexicon and the more quantitative they are. I'm biased for those kinds of data collection practices. So that's one. The other is really over time making sure that the product names are consistent because then you'll do yourself some favors if you can integrate data sets over time. And of course, you want to make sure you have the right checks to make sure that makes sense. And then for us, we're doing a lot of drivers of liking analysis. So making I don't know how painful, how many times it's been so painful trying to link up the product names in sensory versus the product names in consumer. And really sometimes it's your best guess and it's much more reliable if you can at least make sure that the names are consistent.

John: Right. So when the data are captured, making sure that really understand how to understand what's the same, to be able to match things up later. Yeah, definitely needs to be an eye to the future. What about the actual structure of the data? How important do you think that is in terms of because, you know, I mean, I've seen data in all kinds of shapes and sizes coming in sensory in terms of making sure the data are always structured the same way or similarly enough that you could make an entry into pipeline?

Denise: Yeah, absolutely, because particularly with the complexity, it's not so bad with sensory, right? Because it's usually products and then you have the attributes as sort of your column headings. But with consumer data sets, because we ask so many different questions like, you know, like right now we're struggling with how do we format a question design that's like a best, worst scaling, because it's kind of an interesting thing where you actually have a couple of different responses corresponding to product set and then how do you capture that complexity? And so having a consistent shape to the data, particularly on the consumer side, is very powerful. So that's one end of the process. The other end of course, is the storage of it. And so obviously relational databases are well understood across the data space, but I'm seeing a lot this the emerging technology around graph databases. And I don't think you should embrace the technology just because it's cool and shiny. But what I do like about it is we've had to now adjust our data model like 10-20 different times. And it's so fantastic that you don't have to plan out the structure of that data from the get go. And it's super easy to just be like, well, let's add another note here, another relationship there. So I just I find it a very dynamic way to structure your data. So I don't know if there are other people out there who have strong opposite opinions on how the data should be structured from a database perspective. But certainly for our use case, it's been really helpful.

John: Now, I agree that 100 percent, I mean in fact you may have seen that Kyle McNamara, friend of mine. He's the CEO of a company called Graphable. He's been on the show and I definitely am a strong advocate for graph databases and I think that's really excellent. Yeah. I mean it is true the flexibility there is kind of unrivaled and that's a big thing within sensory and consumer science. Our data are very diverse and we always have new sources of information coming in. So actually I think that would lead us to something else we've been talking about or some of the other sources of data that you've been collecting. So what are some of the new exciting sources of data that you see coming in sensory and consumer science?

Denise: Yeah, so I think it's an exciting time in general to do anything in data, but especially for sensory and consumer scientist, right? Because now there are these publicly available, I mean, to be fair, this has been true for a while. There are publicly available data sources where let's say something minimally like understanding the lexicon that consumers are using. You can go in the Amazon product reviews potentially. Sometimes you have Twitter feeds. You can even go and look and read it and see what people are posting about products, right? And so there's just a wealth of sources that you can gather for us. You know, this is an emerging space of a research, but certainly it's very powerful. Like what you can do if you can't figure out a way to kind of cross-reference and marry this diverse set of data. The other emerging space that I've seen is around sort of the neuromarketing space, right? Where beyond just kind of, you know, a survey, there are sort of non-verbal cues that you can you can use. I don’t know how mature the space is, honestly, because the analysis of biometric data and integrating it into more standard practices that's still I haven't seen, I mean, in conferences there's a ton of them, right? But how useful they are in life, I think is something that each company will have to figure out for itself. But it's certainly another kind of input that you could put into the process and see how it helps you understand the consumer better, right?

John: Yeah, definitely. And I would add to that implicit measures, things like reaction time, this kind of measurement. Yeah, actually Michelle Niedziela who is on the show. She's in HCD Research which specialize in a lot of these. And I think that machine learning will help us to understand which variables are important, which are not how they interact. So that’s good. So I would like let's see, maybe we could actually just take a step back because I actually don't know your full kind of origin story as far as how you ended up. So maybe just take a minute, because it is it is funny I do cross paths with a lot of chemists. There's something about chemistry and data science that go really well together. So can you kind of describe your process, you know, we want to talk about how you went from graduate school, you know, to postdoc and then on into GSK. Can you take us through the journey that you followed to end up in consumer science?

Denise: So I actually started off doing something called protein crystallography. So I actually used to go into the lab and actually make stuff. And I found it was a really, really terrible wet lab chemist. So I had to, like, run away and actually do something where I wouldn't cause any damage to the facility, so after a certain point, I met someone at UNC that was doing a lot of simulations of molecules. And so I started working with them. And, you know, I'm showing up at my age and we were doing every some of the set up in Perl. And I wouldn't wish Perl is good for anybody.

John: I've programmed in Fortran so it's okay.

Denise: There you go. A nuclear power plants are still programmed in Fortran. So you have a future there, I guess.

John: Okay, so my backup plan.

Denise: Yeah. And so I ended up doing this postdoc after graduate. So I have some computational papers and then I end up doing a postdoc in a lab that did both experimental and computational methods so I could have my toes in both worlds. So I guess I've always been excited about having a little bit of everything and learning new things. And I think that's what made me excited to join consumer health because I wanted computational chemistry is fantastic and a really interesting space. But the real the focus is always about the chemistry. And after a while, I wanted to try something different. And this is just a great time, I think, to do that sort of thing, because there's loads of new data sets that are coming out, exciting things happening, image analysis, text analysis. I mean, you know, we talked earlier about emerging new data types and just the analysis of language and words has just exploded with these new deep learning models. Right? And so and this is why now we have services like Google Translate or Alexa. I know there's a bit of work there. So it's just I think it's fun to be able to do a lot of different things, which comes full circle to the first topic we were talking about, where, you know, as sensory scientists, this is a great time to kind of hybridize yourself.

John: Yes. Yeah, definitely. Now, I mean, this is the age of the kind of multi-disciplinary approach. Yeah it's definitely is the case that, like, the more experiences that you have that you can kind of bring to your job. I think actually it's the best time really ever to be able to go pulling from multiple disciplines, multiple backgrounds, bringing multiple approaches to bear on problems. So, kind of in that vein, what are some of the exciting opportunities that you see going forward for your kind of consumer science? What are you thinking about in terms of, you know, the future? I know text analytics is something that you mentioned. We talked a little bit about that. So maybe it'd be interesting to, as you look for the next year or two, what are the things you're excited to get into?

Denise: So the other thing that I, this is kind of specific to the OTC space or at least the medicine space. There is something called real world data that is an exciting and emerging area of research. So for folks who are not familiar with health care, often, right, you'll run studies and you'll try to prove out the functional benefits of your product. So if it's coffee, it might be like it gives you a zing or it might have a pretty good flavor profile that is really enjoyable for OTC medicine, it's really about the efficacy, right? And traditionally, you run a clinical trial and you promote the efficacy. And that's what the regulatory agencies like the FDA really want. But more and more, there's this recognition that a controlled clinical trial only gives you one aspect of the consumer or the patient's journey. And by integrating other kinds of data that kind of reflect sort of the real world experience of the patient, then you kind of have a richer picture and you have a new way as well to support efficacy or functional claims on your medicines. So this is the kind of new area of research that's happening both in pharma and now within the OTC space. And the kind of data that is coming out of that research is also diverse. So the traditional real world data set comes from electronic health records and insurance claims data. The challenge with the OTC space is your doctor generally doesn't capture what vitamins you take, whether you're on aspirin or ibuprofen or whatever. Right? So it's kind of a gap for OTC. But there are other types of data that are being captured that are more useful for OTC companies. They're usually things that you kind of have to generate yourself, but there's a lot you can learn about it. So, for example, there's wearable data and I mean, recently we saw Apple watch this, what is it like an app, right, that kind of detects when you have irregular heartbeat? So it's kind of exciting because now your iPhone can tell you if you're having any kind of heart issues. And that I think, is what's exciting for us. It's a bit more about understanding, more focused on consumer research rather than sensory. But what's exciting there is we could potentially have real-time insights about the consumer because the using just survey methods, it's very static, right? You go in, you do a CLT, you do an HUT, and you’re done for the day. But imagine a scenario where you could take snapshots of what the consumer is experiencing, how their behavior is changing. That's super powerful. That opens up insights that previously we couldn't learn.

John: Right. And 5G is just going to accelerate that. Right? As you have Internet everywhere and you've got wearables. Of course, you can have augmented reality potentially play into this, too. Yeah. I mean, I can see situations where actually already doing this, you know, we've talked about the smart speaker platform that Aigora has and we're putting a diary component into that where we want to do is be able to have I mean, the goal here, I think is technology is bringing there's kind of these two worlds, right? There's observational data and then there is, you know, experimentally designed data is collected, you know, in a more intentional way. But then there's the limitations that it's hard to measure something without interfering with it, right?

Denise: Absolutely.

John: In fact, it’s both technically impossible. But, yeah, it's you know, most of the time you're going to measure something. You're going to end up affecting it somehow. And I think technology is allowing more and more these kind of covert measurements to happen. Like you're talking about measurements throughout the day. But I also think technology is helping to bring these two worlds together because like with the diary, if you have someone enrolled in a survey where it's just always happening in their house when they're in their house, they're supposed to tell their smart speaker, you know, I'm not going to say the word because there are smart speaker on here. A smart speaker record that I ate a banana. Smart speaker record that I had some cookies or whatever. But the thing is, the smart speaker is looking out for particular things you might say. And then when you, so say, smart speaker record that I took my aspirin or whatever, might ask some questions, then say, well, excuse me, do you mind if I ask a few questions? And so you might get so now you have this blending of the kind of observational measurements to with more designed because under some circumstances you might ask follow up questions. Very interesting to think about combining that with wearables, because if you know, suppose that you're interviewing someone, a smart speaker, but you also have their wearable data from that moment. Right? How are they physiologically responding when they were drinking the thing? Yeah, it is an awesome time, honestly. Alright. Well, let's just, I do want to know a little bit, and I don't want to get you into this kind of trouble, but I'm curious how you because you've kind of been in the two worlds, GSK, right? You started off on the pharma side. Now you're on consumer science side. So how do you see the approaches differing? And are there things like what do you see that the two sides can learn from each other? What are some opportunities there for?

Denise: Yeah, that's a good question, actually. I think from the pharma side, they just have longer timelines. So they have the breathing space to really think about the science very deeply. And so when it comes to thinking about how and I'm going to loosely call them functional benefits, what they're really about understanding how biology and chemistry come together, they have years to really think about that very deeply. So if you're looking at a paper and you see a pharma company on that, you do know that there's a high degree of robustness in the data that underpins that paper. With consumer health our timelines are just so short and we don't have the breathing room to do that. And so not that the science isn't robust, but it’s not something like we don't have like 10 years, which is what it takes to launch a drug, to really think that deeply and understand every nuance of the biology. But what we were really great at is understanding the consumer and understanding how a particular therapy or drug or medicine or even like even like, let's say telemedicine is kind of another emerging area of health care. We can understand very deeply how you find a solution that really meets the consumer or the patient's need. I think that we're much closer in our relationship with them.

John: Interesting. So you would say there's more of like psychology that the psychological piece is kind of....

Denise: Yeah. So for us, let's say if you're thinking about a kid's pain medicine, will think very deeply about like, what does a child want and how can we make sure that the medicine is something that they'll take and so that it'll really make them feel better. On the pharma side, they'll think about that, but it's more like let's take the box. You know, what they really want is to understand? Is this the right kind of molecule for the child? You know, how is it affecting their biology and learning about that very deeply. And the kind of sensorial aspects of it are important, but not the first thing that they're going to start thinking about. Does that kind of make sense?

John: Yeah. Makes perfect sense. If the medicine doesn't work, then the experience is less important.

Denise: Exactly.

John: Right. Okay. Now we are actually almost out of time. So I do want to ask one last question, you know, because I don't really know what the right answer to this question. People ask about this all the time, which is do I need to learn to code? Right? And to what extent do you think that sensory scientists right now really should be taking courses in data science, learning python, learning R, you know, what do you think are going to actually be the requirements going forward for young sensory scientists in terms of coding skills?

Denise: Yeah, I think it'll become easier and easier to write up code because the packages are getting progressively easier to understand. And so as a sensory scientist, I would say, of course, learning the sensory science, make sure your understanding of statistics and mathematics is robust, but you do have to learn how to code because otherwise you're going to be limited to the tools that either you pay for or the kind of modules that exist. Right? And so you open up really a range of different tools for yourself. There's also you mentioned the automation. So do you really want to spend hours and hours making the perfect PowerPoint when you click a button and then the PowerPoint slides are made for you automatically in the same way? And of course, that automation lends itself to reproducible research and analysis. So you cannot get away from coding and it doesn't matter which language you decide to put up, first R or python. It doesn't matter. It just you do have to get savvy enough to at least be able to write up some scripts for yourself. There are publicly available like repositories start to learn, to do simple things quickly. So it's not like this is I mean, the advantage is there's so much resource out there these days. It's not that hard to learn.

John: Yeah, I agree. That's right. I actually find it with my clients. Sometimes the biggest hurdle is simply the belief that they can do it because they actually can. And a lot of times when people get into coding after a few weeks, they're like, oh, this is a lot easier than I thought. I didn't realize that you did this and I can do it. So, yeah. I think leap of faith is require.

Denise: Even in my own family. So my husband is a chemist as well. We are a family of chemists and two years ago he was like, I want to change my career. And so he pivoted and took some Coursera courses. And now he's also in the data space leading a team on the pharma side. Yeah, so but he likes R, I tend to use Python more.

John: If I ever get any free time, I'm going to go back. Four years ago I learned Python and I didn't, I hardly used it because there are just so many tools in R for sensory, you know, that its but I think there are some advantages to Python. So one of these days we will see what I don't know how many keep Aigora going during this time, but I will start studying Python. I would like to learn more Python for sure. So alright, Denise, if someone wanted to get in touch, I mean this has been great. I suppose someone wants to apply for a job in GSK, or have some question for you, this might connect, what's the best way for them to get in touch with you?

Denise: Probably through LinkedIn. So just look me up at Denise Pohlhaus. I don't know the exact website, but Google will help you. Just look for me online and reach out to me.

John: Yeah, and we'll put the link in the show notes for the podcast so people can just click on that link. Alright, so the last question we always ask is, what advice do you have? Maybe we already talked about it for young sensory consumer scientist. What should they be thinking about over the next couple of years? Someone just graduated that join GSK, what advice would you give them?

Denise: Yeah, I would, of course, pick up some coding skills if you don't have it, but never lose track of your ability to interpret the data and communicate it. Don't be overwhelmed and don't feel like, oh, I'm never going to be able to do this, because at the end of the day, what and I see this also with the cop chemists, the most successful cop chemists are people who not just can spin up some code and run some mathematical calculation, but they can interpret the data very deeply and communicate that very convincingly. And that's a hard skill to have. So don't forget and don't kind of discount those skills that you learn as you become a sensory scientist, because they will also make you a more successful data person.

John: Communication. Yeah, that's interesting. Yeah. I mean one thing I would add to that on that point, you know, at the intersection of what you said, encoding would be data visualization because you can really think about how to visualize your data in a way that tells a story. It's really compelling.

Denise: Even if it's just a line chart, you never know how powerful and exciting that may be for your end user.

John: Yeah, definitely. Okay, Denise, this has been great. Thank you so much. Anything else you want to say?

Denise: No. Thank you, John. This is my first podcast, so it's been a very exciting and positive experience.

John: Oh, well, congratulations. Yeah, it's great. You've been an excellent guest, so thank you.

Denise: Thanks.