LabOps_Tommaso Mansi: this mp3 audio file was automatically transcribed by Sonix with the best speech-to-text algorithms. This transcript may contain errors.
Kerri Anderson:
By building a platform to share challenges, thoughts from leaders, and network together, the LabOps Leadership Podcast is elevating LabOps professionals as well as the industry as a whole.
Samantha Black:
With the intent of unlocking the power of LabOps, we deliver unique insights to execute the mission at hand, to standardize LabOps, and empower LabOps leaders.
Kerri Anderson:
I’m Kerri Anderson.
Samantha Black:
And I’m Samantha Black. Welcome to the LabOps Leadership podcast.
Samantha Black:
Today, we’re here with Tomasso Mansi, who is VP of Artificial Intelligence, Machine Learning, and Digital Health at Janssen R&D. Thank you so much for joining us today, Tomasso.
Tommaso Mansi:
Oh, thank you so much. Very nice to be here. It’s a great honor to basically share our experience in AI with you, and I look forward to the discussion. Thank you.
Samantha Black:
Thank you. Well, let’s kick it off, and if you just want to tell us a little bit about how you got to where you are today and just a little bit more about your background.
Tommaso Mansi:
Sure, so as you said, right, today I’m working at Janssen R&D in the data science team, right, focusing on development of AI solutions to help our programs. But before arriving where I am today, actually, I had the chance to evolve in healthcare, but coming from a technical side. So everything started when I did my master’s at Montréal Neurological Institute. So my background is in computer science and telecommunication engineering, but then I had the chance to start working on neuroimaging data for actually pediatric use cases, so patients suffering from epilepsy, and it felt so, a great sense of purpose, right, that you can actually develop solutions that can help patients. So it’s where somewhat I took the virus of medtech, if you want, then I had an experience in industry and finally switched to do a PhD in actually biomedical engineering. So doing some medical imaging, AI development, modeling and simulations, digital twin, and that PhD was in collaboration with Siemens Healthineers, which just after finishing the PhD, I had the honor to join. And so prior to being at Janssen, I worked for Siemens Healthineers for 11 years, and at that time we were developing AI solutions to support basically physicians to do their job, and my focus was mostly on image-guided therapy, so how you could use AI imaging, real-time sensing to help the surgeons and physicians better treat patients. And then I had this wonderful opportunity to actually join the Janssen R&D data science. You know, pharma is a therapy by excellence, right, so the first thing you give to treat a patient is the drug. And I’ve always, you know, tended to go towards therapy because if you fix and if you treat a patient, at the end, the patient is happy, right? And so I said, why not join? And really, it’s amazing what actually AI machine learning data science can bring to the pharma world and the impact it can have to the patients that we serve. And so that’s been an exciting journey and I’m happy to discuss more with you, what can be, what is possible?
Samantha Black:
Yeah, I’m very curious. You know, it’s a very big umbrella that you’re describing. So can you tell us a little bit more about what you’re actually working on and what technology you’re developing there?
Tommaso Mansi:
Sure, I think it’s, so if we step back, right, and you see what is currently happening in healthcare, actually, as care industry is experiencing what is happening also in other industries, with multiplication and exponential modification of data sets, right, you have progress in sensing, progressing in algorithms, print progress in compute, which actually makes things possible today that we could not even do five years ago. And in healthcare, this is you know, you see it every day, right, new technologies to sense what the cell can do, new machines, new imaging that is more and more precise. The data is being digitalized and managed into any hospitals, right? So all this creates actually an ecosystem that allows us to develop solutions, right, to basically develop novel treatments for the patients, and this is what we recognize in Janssen R&D, right? So this, and Janssen in general, so it’s this power of this combination of increased biological knowledge, this is how it works and the data, and then the technology, right? So this convergence that can bring so much, right? So how do we do that more precisely? Like I told you, so we really believe on the power of data to give the insights right, on diseases, right? So leveraging large Biobank data sets to try to identify subtypes of diseases, for instance, I mean, you may call atopic disease like mood disorders, for instance, right? But actually, there are several etiologies below them and then each etiology might actually result in different mechanisms that you can target and potentially new drugs, right? So on the clinical trials themselves, right? So by bringing data set in the operationalization, you can make it more efficient, potentially shorten the time, increasing the specificity to, again, help getting the insight on the efficacy and potential side effect of the drug. And then, you know, precision medicine is like, precision medicine is actually happening, right? We are developing treatments for more and more targeted diseases, right? And now the key is to identify that disease, but then also screen these patients, right? And for that, you need advanced technologies and sensing and algorithms to identify these patients. And so this is what we do, and particularly, in my team, we focus mostly on AI algorithm developments and solutions, and development of solutions, right, to really look at end-to-end from the early understanding of the disease, all the way to help in clinical trials, right, with advanced solutions ever, for instance, in medical imaging or also in digital health. And it’s really a great honor, right, to be part of a great team, 100 scientists who are dedicated to basically help, to develop a drug and bring them faster to the patients.
Kerri Anderson:
Yeah, I think it’s incredible you’re able to target in that way, and the technology has just changed so much in the last few years, but as we know, with rapid growth, we often encounter many struggles. What are some of the struggles that you’ve encountered during this time and what were some possible solutions you came up with?
Tommaso Mansi:
That’s an excellent question, right? It depends where you look at, right? So one of the struggles is can we, so, you know, I mean, there is the success rate of a new drug throughout the development process is not as high as we would like it to be, right? So you have many clinical trials that fail and so forth. So the key question is how can you use data to actually increase that success rate, which basically means do better decisions earlier in the pipeline? Can you, for instance, even before going to clinical, predict potential toxicity of a drug? Because if you can predict that the drug will have off-target toxicity, right, then you stop right away, or also potentially predicting efficacy, right? So you have the preclinical experiment that of course are being done, but actually, there is a trend to push that decision, push that prediction earlier and earlier thanks to actually high throughput screenings, for instance, where you have the combination of automation and robotics with new sensing technologies and cell lines, organoids, and all these things, plus the massive amount of data sets that you have, right, to, what you can generate to actually make these predictions, but then also link it with real-world data that is out there, and these large biobanks, for instance, that exist, right, for instance, the UK Biobank. So can you, so more insight you have earlier to help a decision in order to overcome this first struggle, which is the success rate of the trials. Another one is more on the trial site is that, is on the diversity, equity, and inclusion, right? When you develop a therapy, right, you want it to be, to work on the population that is in the world, right? And for that, it’s super important that when you develop the therapy, right, you do it with the population in mind, that you don’t only go in one specific site, right, and then overlook over populations, right, who could, would, we want them to benefit from that therapy, but because we cannot involve them during the clinical trial when it’s restrictive. So this is very important for us. We really, really want, we really have this in our heart and using data actually to identify, for instance, what would be the best sites in the world such that we can effectively test and try our drugs in the population we target and have a representation of everyone, right? And here again, data helps so you don’t go anymore to your, let’s say your preferred vendors. I mean, we still go with them of course, but we do a strategic decision on where to go such that the population of the clinical trial is representative of the population who want to benefit of the drug we develop. So site selection, patient identification, right, developing AI solutions to identify the patients early on to basically give them the opportunity they are willing to enroll into a clinical trial. And then finally, the reality is that you may want to participate to a clinical trial to benefit from the latest therapies that exist, but it might be difficult because maybe the hospital that is doing that is at 100 miles away, right? So how can we still give that opportunity to patients, right? And so this is a power of actually what we call decentralized clinical trials. So can we use digital technologies to actually bring, quote-unquote, bring the trial at home, right? So use wearables, for instance, to quantify the state of the patient and send that information back to the PI. You can do the analysis, and instead of asking the patient to go every week to the hospital when to go less often, right, so that would actually increase the access to the latest therapy to these patients. So you see so this is data and AI can help in all these aspects, and this is very exciting because they have the potential to provide solutions to the struggles that we all face in the industry.
Kerri Anderson:
Yeah, I think it’s incredible that the data and AI is able to help with that, especially the inclusion factor. It’s really important today.
Tommaso Mansi:
Yeah, this is a huge, that’s a huge opportunity that we have to, right, to do it.
Samantha Black:
So I’ll jump in here. I think this is incredible and I think data is at the center of it, and whenever we talk about data, you know, I think the first thing that comes to my mind is data integrity and how do you trust the data that’s coming out of the lab? And so from our audience, I think people who work in operations, this is a big point for them. They’re really working to make sure that the data coming out of the lab can be trusted. So I would be interested to know from your point of view, you know, how that impacts your work and how you are ensuring that the data that you’re using is validated and is, correct is not the right word, but is to be trusted and is the right data to use?
Tommaso Mansi:
That’s an excellent question, right? So there is a saying in AI that garbage in, garbage out, it’s not, basically, it’s fundamental to have proper data governance, right? And in data governance is not only the data quality but also everything around data like privacy. It has to be linked, you can only use the data for what you have been granted for, you know, traceability, it to be within your, regarding data quality, right? So that’s an excellent point. And we put in place processes, so first, we start basically, by the question, what do we want to answer, right? What is the problem we are trying to solve? So this is fundamental, right? And then based on that, we ask the question, okay, what data is available and what data can be generated? So when it’s on the technical side, for instance, you may have actually even the possibility to start using automation and large-scale robot and robotic experiments that will actually decrease the variability of the experiment by basically going large scale, but a very thorough process. When it’s on the clinical side, of course, we have to set up quality control and checkmarks, right? So on the data that you obtain, usually you have a well-defined protocol, what type of imaging, what type of lab test that you acquire. Then you have data annotation. So one is, for instance, the raw data, and the other one is the annotation. The annotation means you basically assign a label to the data that you have, and that label is used to train the AI system, right? Here, actually, having very, very good annotations is even more important than having very, very good data, right? Because the reality is that when you want, for instance, to deploy an algorithm in the real world, you need to train your algorithms on the data that you will see in the real world, so sometimes you may want to have datasets that are of real-world quality, right? But on the annotations that you use to train the algorithm, this is basically the teacher of the algorithms, so the teacher has to be right. And here, again, you have different ways. So if, for instance, if the label is a molecular test, of course, you make sure that there are proper protocols and so on and so forth that are followed throughout the different patients, for instance. But if it’s something that somebody has to label right, then again, we need to rely on experts. We did that, the lifetime to do the labeling. The reality is that sometimes one expert is not enough, you need to have two or three to do the, each one do the labeling. They will not, in some of the cases, they will not agree. Bring them in a room, come with a, with an agreement, right, and then use that as a ground truth for the AI development. And then the other thing is that you will have also data from other modalities, right? We’ll have imaging, EHR, lab test, this, and so here’s the key, and it’s connectivity, right? How do you link all this information, right, in order to then when you need that information to access the data? And here in Janssen R&D, we developed an infrastructure called Medi-I, actually to help us throughout the entire process, right, from the governance to data cataloging to linking between different modalities to the, linking with the annotation such that the scientists, right, can just log, look for the data we need to answer the question they have, and then confidently use, if they are allowed to do it based on on the governance, right, so the system tells you okay you can or you cannot, then confidently use the data for their algorithm development. Back to the lab, right, I think where there is a huge opportunity, but, please keep me honest here is, on the automation, right? When you look at what is happening, right, is this convergence of robotics, AI, and Wet Lab? This is so exciting. I mean, it’s like it’s amazing what we can do today, and five years ago we could not even do, right, by combining computer vision, combining robotics, combining novel biological processes, right, that allow, to basically alter the cells and so on, so forth, we can generate actually data at scale. And because it’s, you bring more and more automation, right, we’ve, of course, the human in the loop, but this interaction between human and automation actually helps scale, better, to standardize and this helps the AI development down the line. So this, I think the centaur, right, in AI, we have this centaur concept, which is basically the joint partnership between human and automation, robotics and the AI, right? This is actually super exciting, right? Because you take the best of the two to achieve breakthrough that we don’t even know what we’re going to do.
Samantha Black:
So cool, like it’s, it seems like the, very futuristic but it’s right around the corner. Like you said, it’s already happening in some ways. So I think it’s very exciting and I think, one thing for me is people are still central to that, I think. And I think whenever anybody says AI, they think, oh, something, just computers all on their own, like it’s going to run the lab, and I think what I’m hearing from you is that’s like never going to be the case because you need people behind the scenes, you need people, LabOps, people in the labs, you need scientists, you need data scientists behind the scenes, making sure that it’s doing the job that it needs to get done. So I think that’s a really important point, and I think the steering of it, the steering of that is fascinating to me because you can’t remove people from the equation. I think that is encouraging, I would hope, to many of our listeners who are, you know, there’s so much possibility in AI, but, you know, maybe there’s also opportunity for LabOps or anybody in the healthcare sector to really drive this forward. So I think it could be seen as an opportunity rather than a scary thing that humans are going to get cut out of the process, and I think that’s, you know, I just.
Tommaso Mansi:
I fully agree, and, you know, I was before in the radiology and when AI started in radiology already five or six years ago, there were lots of articles, right, “should even students take radiology as a specialty?,” right, in the news. And the reality, you know, is that, no. I mean, I think, if we shoot, right, is that AI will not replace them, and they were, actually, even, there were people saying radiologist plus AI, right, will replace AI-only of course, but also those who don’t want to use AI and just stick to manual stuff. So it’s really a combination of the two, the synergy that can bring a lot of value. And we see it across many industries, right? And it’s where you take the ingenuity of humans combined with the scale that automation and AI can bring, right, where insights will be generated. And so that’s why on the lab I think you see a lot of, you see, for instance, companies developing, actually, with integrated infrastructure where you have an experiment that is run at scale, and then the CE lab or the scientist or whoever gets result to the analysis, try to understand and potentially make another hypothesis and then regenerate again. But then sometimes you always need the quality control, of course, right? And also you need also to invent, right, as new discoveries come, and this comes also from human, right? So it’s really the symbiosis that is very exciting, but that’s my personal opinion, right?
Kerri Anderson:
So yeah, I couldn’t agree more. I mean, it’s that combination of automation and people that will be driving us forward. But, so you’ve had this incredible career and it’s been fascinating to hear about. I’m sure you have a lot of advice to offer our listeners. What’s the biggest lesson you’ve learned so far in your career?
Tommaso Mansi:
That’s a tough one. Just be curious. Don’t hesitate to go beyond your comfort zone and be humble and learn. I think what is fascinating, it’s connecting dots across fields that on the surface or are fundamentally different. But when you, sometimes you have cross-pollination that happens, right? So something that you do in one field can be transferred to the other one and recently comes through curiosity and learning what the other person does and you know, bring, there is no stupid idea. So that would be, you know, a lesson learned. So many times I had the chance to actually talk with people doing something completely different and then come up, hey, actually maybe very something, you know? So shed another light to a problem, is fascinating. And then when you, when it becomes technical, right, and when you start talking like more on the methodology technology side and then go further down on the mathematics, right, then there is even more common denominator. So sometimes you might be using similar approaches in fields that are completely different, but the mathematical theories actually are the same.
Samantha Black:
So if you don’t have to reinvent the wheel, don’t do that, right?
Tommaso Mansi:
But this is one, of course, right, but the other thing is that sometimes in your field you may have a problem with solution has been already developed in another field, but it’s just that you don’t know because of, you know, we have specialized. When somebody just comes up and then say, oh, this has worked for me, then you actually solve your problem, right? I think I can give an example. For instance, I remember we developed an algorithm. We were working to develop an algorithm to predict the electrophysiology in the heart, right, to simulate, basically, create a patient-specific model of the cardiac electrophysiology, and it takes time to compute that, right? It’s a complex model that takes into account what happens in the cells, and blah, blah, blah. In a completely separate world, in computational fluid dynamics, that is being used in other industries and so on and so forth, we actually developed a method that allows you to do very fast computations, right, because we had this need. Now, one is like fluid dynamics, right? So air and water and stuff. One is electric, electricity if you want, but when we were chatting with a colleague and when we started to deep dive, and when you realize that actually, the equations are different, but the fundamentals are connected. And when you say, why don’t we try, and actu1ally, it worked like this, and we transported basically that technology to the other field, right, and we could now actually do the computation very fast. And so this kind of, you know, this cross-pollination is, at least on the technical side, it’s something that is very exciting, and that’s why being curious of what is being done in other areas, you always learn something so that can be beneficial.
Samantha Black:
Yeah, I think, I mean, that’s where data is a connector between everything. You know, it’s not biology maybe, but you know, or it’s a different industry or different type of medicine, but data is underlying all of it. So I think there’s a lot of lessons to be learned, no matter what field you’re in, so awesome. Well, Tommaso, the last question I have for you is if people want to follow your work, if people want to learn more about what you’re doing, where might somebody find you and how can they connect with you?
Tommaso Mansi:
Sure, I think the first go-to place would be going into our Janssen.com website or global channels on LinkedIn, Instagram, where we publish basically the latest news of the science that is being developed. Yeah, I think that would be a good entry point. You can look at the profile.
Samantha Black:
Awesome. Great, well, thank you so much for joining us today. This has been a really awesome conversation, super interesting, and I know that our listeners are going to find a lot of value in it, so thank you.
Tommaso Mansi:
Thank you so much. It’s been a great pleasure, and thanks again for the honor to, that you gave us to be part of this discussion.
Samantha Black:
Great, thanks.
Kerri Anderson:
Thank you for tuning in to this episode of the LabOps Leadership Podcast. We hope you enjoyed today’s guest.
Samantha Black:
For show notes, resources, and more information about LabOps Unite, please visit us at LabOps.Community/Podcast. This show is powered by Elemental Machines.
Sonix has many features that you’d love including automated translation, powerful integrations and APIs, advanced search, upload many different filetypes, and easily transcribe your Zoom meetings. Try Sonix for free today.
