AI's Rx Revolution: Transforming Pharma

Get ready for a mind-blowing episode of 'AI Health Heroes' as we decode 'AI's Rx Revolution: Transforming Pharma' with the brilliant Dr. Xiong (Sean) Liu from Novartis!

We'll prescribe a dose of innovation, uncovering fascinating insights on AI-powered drug discovery, ethical AI in healthcare, and data-driven patient care that'll make your neurons dance with excitement.

Prepare to have your synapses fired up as we navigate the world of machine learning in pharma, NLP in healthcare, and AI-driven diversity initiatives, discovering how they're curing the healthcare industry's toughest challenges.

With Dr. Liu's expertise in AI and data science, you'll hear inspiring stories from the frontlines of pharmaceutical innovation, leaving you dizzy with possibilities.

By the end, you'll be inoculated with these game-changing takeaways:
1. How to integrate AI into your health research
2. Strategies for balancing innovation and patient safety
3. Tips for fostering collaboration in AI and data science teams

Don't miss this appointment with the future! Tune in now, subscribe, and join our health tech revolution. Together, let's code a healthier tomorrow!

🎙 AI50 Podcast

📹 Want to receive our videos faster? SUBSCRIBE to our channel!

👉 Visit our AI50 website

👉 Schedule a demo

📰 Receive our weekly newsletter

👉 Follow Hanh Brown on LinkedIn

🏛 Follow AI50 Business Page

Find Sean on LinkedIn: https://www.linkedin.com/in/xiong-liu/

Transcript

Sean: 00:00:05
I think nowadays we are in a golden age. This will potentially cause a paradigm shift in drug discovery and development. And also there's not just an efficiency gain, but also a new insight gain. So for example, when we're designing our new molecules, now we can use those kind of generative AI methods to design brand new molecules with designed properties that never existed before.

Hanh: 00:00:33
Hello, I'm Hanh Brown. Welcome to AI50, where we harness the power of AI to create innovative solutions for the 50 plus demographic and their loved ones. Our mission is to develop cutting edge language model applications to cater to the unique needs and preferences of older adults and their parents and grandparents. Whether you're part of the 50 plus community or have aging family members, Our goal is to ensure that everyone benefits from advancements in the AI technology. So join us as we explore AI's

Hanh: 00:01:08
potential to revolutionize the way we live, learn, and connect. So I'm thrilled today to introduce our guest, Dr. (Sean) Liu, a pioneering leader in AI and data science at Novartis, a global healthcare powerhouse. Well, imagine struggling with a complex, life altering disease, desperate for effective treatments and hope for the future. Well, that's where Dr. Liu's groundbreaking work comes in. With his expertise in AI and data science, he's revolutionizing pharmaceutical

Hanh: 00:01:47
research and development, accelerating the discovery of innovative medicines across multiple therapeutic areas. Including cancer, cardiovascular diseases, and rare disorders. So today, we'll dive deep into Dr. Liu's inspiring journey, the challenges of integrating AI into pharmaceutical research, and the ethical considerations surrounding these cutting edge technologies. We'll also explore the lessons he's learned from his SBIR awarded data mining research. And his vision for the future of AI

Hanh: 00:02:20
in patient, in shaping patient care. So, Dr. Liu, welcome to the show. Thank you, HaHanhthank you for having me. Yeah. Well, thank you. Thank you so much. So, can you share something about yourself that maybe not too many people might know?

Sean: 00:02:41
Yeah. I'm glad to talk about AI Infirmary, uh, my personal experiences and my views. Uh, as a standard disclaimers abuse mine, not the company. So, uh, I have background in engineering and computer science. I got my Ph. D. Information science from the University of Pittsburgh. So when I was looking my Ph. D. Research topics, I look at a long list. And eventually bioinformatics

Sean: 00:03:07
touched me because from early age, I know that biology and medicine could improve human health. But I thought it's mainly conducted through experiments. While I am not very hands on and not a handyman. So I thought it's kind of a little bit challenging for me. But then I realized that, uh, using compute computing computational methods, we could also unlock a lot of insights in biology and medicine. And suddenly that widened my eyes. So I decided to I want to focus on this

Sean: 00:03:40
because it can give my work more grand meaning of which is improving health. And after that, I went to postdoctoral training in bioinformatics at Johns Hopkins University, so where I learned gene, Uh, regulation and, uh, how genes are expressed in different human tissues and that relevance to, uh, uh, human diseases. Um, and then, uh, I went to a small ai, uh, tech company called, uh, intelligent Automation in Maryland, DC area. Uh, so, uh, it is, it, it was there that I transitioned from, uh, a student, uh, a post-doctor fellow to an, uh,

Sean: 00:04:21
independent principal investigator. Um, so I, uh. Uh, what was applying, uh, um, the small business innovation of grounds, uh, with the government government agencies and also conduct research in, uh, data mining, but also with applications in biomedical, uh, uh, area. Then, uh, I thought I could, um, continue doing this until I got an opportunity to join Eli Lilly and the company. Uh, so my years at Lilly, uh, have, uh, two parts. Uh, the first is with the, uh, research IT and, um, Research,

Sean: 00:05:00
uh, informatics, uh, department. And second part is, uh, with the, uh, enterprise level, uh, data science center. So, uh, and the, um, and this happened, uh, coincide with the, um, the uprising of AI and deep learning. That's around, uh, 20, uh, 17, 20 18. Uh, because, uh, AI was, uh, catching. Everybody's attention. Every industry and leaders, they want to invest and see how that improves their productivity. Uh, and I was glad I have. I was being able to witness the whole transition process that how is getting,

Sean: 00:05:38
um, um, prevalent and how people are using that to show its value and benefits to, um, health care to patients to a lot of other, uh, important social issues. Awesome. Yeah. And then, uh, my, uh, now I'm, uh, with, uh, Novartis and also, uh, working on data and AI to, uh, enhance, uh, drug discovery and clinical trials. Uh, so I'm, uh, quite happy I have the opportunity to use data and AI to, uh, help the industry and, uh, eventually, uh, patients and healthcare.

Hanh: 00:06:12
Absolutely. Such important work. And I appreciate it. Your work that you do now. A recent study by Accenture suggests that AI has a potential to revolutionize the pharmaceutical industry with an estimated 150 billion in savings by 2026. So what inspired you to pursue AI and data science? And I know you've touched on it previously. Was there a particular experience or realization that motivated you?

Sean: 00:06:42
Yeah. Um, so, uh, interestingly, uh, I have been in the whole, uh, um, journey that, uh, how AI, you know, got, uh, revived, uh, and, uh, now got, uh, you know, such a key component in R& D. Uh, you know, in the, uh, 1980s, uh, AI, uh, there was a movement in AI, uh, especially, uh, in the robotics. And then, uh, when it goes to 2000s and then that's when the Internet and becomes popular. So a lot of, um, people still do, uh, machine learning this kind of research. But, uh, at that moment, uh, the data

Sean: 00:07:17
maybe is, uh, not that big enough. And also the computing power is not that big enough. And probably applications were not abundant at that time. And people usually call that data mining. Thanks. And text mining, uh, image processing, image analytics. And people really call that AI. Maybe people call that a machine learning, but the AI itself was not frequently used. And then when we get in the, um, around the, uh, the early, uh, 2010s, and then the rise of deep learning image processing

Sean: 00:07:51
then suddenly ignite the revolution. And then when we get into the middle of 2010s, then, um, the rise of deep learning and also especially the introduction of the transformer models that totally changed the whole game. Previously, we talked about AI, it's about, it's more about traditional models, about, uh, Logistic regression decision trees. But after the introduction of the transformer and those deep learning models, then it constantly means a different thing. So it means like more computing power

Sean: 00:08:27
and more better pattern recognizer from a gigantic amount of data. To enable insights, discovery to improve efficiency and also even introduce the shift of paradigm. So in the farmer, what inspires me is that there is being a long decades of challenge. We all know that people often see that it takes over 10 years and a billion dollars to launch a new medicine. And now the numbers have been updated. People have been saying it actually takes more than 13 years and the 2. 8 billion to launch medicine.

Sean: 00:09:07
And the success rate is relatively low, is around 10 percent from the beginning to launch. Then people have been thinking how we could improve the R and D. Process the productivity in farmer. Um, there are many people have been looking at from many different perspectives from experimental perspectives and also computing perspectives. And now is the rise of AI. Obviously, we all think, uh, you know how I could improve. So in my view, I could have every

Sean: 00:09:36
aspect of the pharmaceutical R and D. Uh, including the early discovery, clinical trials and real world evidence. So, um, um, yeah, so there's a lot of, uh, uh, contents to cover, but let me just, uh, elaborate a little bit. So, for example, in early discovery. Very important task is about finding targets, designing molecules, and so that we can move to the clinical trials and testing on patients to see the safety and efficacy. So previously, when people are finding the novel targets, they usually go to the literature and also doing bioinformatics

Sean: 00:10:17
analysis of gene expression data, etc. for your attention. to manually put together all kinds of evidence. They can make a decision so you can think that, you know, PubMed has millions of articles and it's really hard for people to digest them and then a new content keeps popping up. So with the natural language processing, it is possible to digest this huge literature and generate precise gene disease and drug relationships so that people can navigate among this kind of information networks

Sean: 00:10:56
to tailor to their hypothesis generation and therapeutic needs. Uh, and also, um, Uh, we'd go. Uh, and also there's, uh, not just, uh, uh, efficiency gain, but also, uh, in a new insight gain. So, for example, um, when we designing our new molecules, uh, now we can use, uh, uh, those kind of, uh, generative AI methods, uh, to design brand new molecules, uh, these, uh, desired properties, uh, that never existed before. Um, and also, um, more recently. Some AI based biotech companies are pushing the, um, AI design molecules

Sean: 00:11:36
against AI predicted targets into clinical trials, and it's been going on quite well, and some molecules have already entered into a phase one or phase two clinical trials. And, uh, if successful, this will prove that it is possible to use data and AI, uh, to reduce costs and budget and speed up the efficiency in pharmaceutical, uh, development. And this is, um, and any of this is successful and more companies, uh, will do the same. And this will potentially cause a paradigm shift, uh, in drug

Sean: 00:12:10
discovery and development. So these are all the exciting things that, uh, motivate me. Um, to come up, work and, uh, every day and communicate with people and, uh, also, uh, make my own, uh, contribution to the field. Yeah,

Hanh: 00:12:23
It's a, it's a great time to be doing work in all regards, you know, in the realm of AI, uh, gosh, customer service, productivity, creativity. Life long learning. So I concur with everything you're saying. Now, integrating AI and machine learning into pharmaceutical research can present numerous challenges. So what have been some of the biggest challenges? And how have you approached and overcoming them?

Sean: 00:12:56
Yeah, um, That's a great question. So definitely there are many, many challenges. So for me, I like to categorize them into two parts. One is about the technology stack. The second is about business use cases and applications. So for the business stack, if we further break them down, it includes data algorithms and platforms. You know, in order for AI to work, Uh, we definitely need a lot of data to cover the knowledge space because, uh, you know, these are the mathematical

Sean: 00:13:27
and, uh, statistical algorithms. They need data to learn what is the, uh, the popular, what is the sample space look like? So, so that way it can generalize into a model and predict new things, given a new inputs. So, um, the data are challenge is obvious, right? I mean, data are desperate and, uh, uh, distributed across different, uh, areas. And also it could, um, you know, coming from, uh, inside, uh, different organizations and it could, uh, coming from the, uh,

Sean: 00:13:57
the public, uh, the open space. Now, how do we harmonize all those datas in a standard format that can fit into, uh, AI models? That's definitely a big challenge, right? Uh, and when it goes to, uh, algorithms. You know, we are all motivated that I can do so many things, but, um, the reality is that, uh, no, uh, no algorithms is 100%. We need a lot of, uh, training, validation and testing to see the performance. We always need a new algorithms to improve the performance. So that way we can make models more practical.

Sean: 00:14:29
in reality. And when it goes to platforms, there are many computing platforms supporting the data storage, supporting the GPU computing, model training, testing and also deployment. So definitely there's a lot of challenges there. Um, and when it goes to the assuming we have worked hard and have all the things in place. The next thing, how can we Turns this, uh, uh, the whole technology stack into, uh, innovative medicines to patients, uh, to help with patients.

Sean: 00:14:58
Then this goes to, uh, very different, uh, disease areas and, uh, the unmet medical needs, uh, and how we define the priorities, uh, what kind of AI models we want to do first. So there's a lot of challenges and concerns there. So, um, yeah, but the, uh, you know, accompanying the challenge also opportunities and excitement. So I think I'm glad to embrace both challenge and opportunities. Yeah.

Hanh: 00:15:24
So now, uh, do you use AI systems such as Azure AI and Fabric to ingest the data or do you have your own at Novartis?

Sean: 00:15:35
Yeah, that's good question. Uh, so, um, you know, um, nowadays, uh, the farmer, uh, industry, now they have a lot of collaborations with those tech companies. So, uh, so it is not uncommon that we use Azure. We use AWS, et cetera. Um, but I would say I would like to embrace the idea of agility. Right? Uh, because many times, uh, our new goal is to build a models useful for decision making, uh, in making new medicines in helping patients.

Sean: 00:16:06
Right. So, um, Many times we start, uh, from the, uh, exploration, uh, at that stage, it does not need a lot of, um, you know, rigorous, uh, computing power, uh, et cetera. We can even, uh, start with, uh, on h HPCs, right? So, uh, so, so for me, um, then, um, I would say nowadays the, uh, the choices of infrastructure, uh, uh, is a lot right? Then, uh, then it's really, uh, depends on the business needs and the cost, budget, uh, et cetera.

Hanh: 00:16:35
Very true. I'm finding out that many companies, whether they're small, midsize or, you know, fortune 500, they're looking for AI systems and not just models because to ultimate, to reap the full benefits of AI is to have an ecosystem that speak to each other, right? So that we don't have various plugins and various operating systems. So, now as AI becomes more prevalent in healthcare, ethical considerations become increasingly important. Now, what do you see as the key ethical considerations surrounding

Hanh: 00:17:10
the use of AI in healthcare? And how do we ensure that these technologies are developed and used responsibly?

Sean: 00:17:19
Yeah, that's an excellent topic and I have a lot of fundamental issues there. So, um, So to me, I think if we think about from the patient's perspective, right, there's a lot of a patient privacy issue. Uh, there is the when it goes to a modeling, there's a buyers that could be biased in data and models, right? And also when we go to the clinical studies, uh, there's also a bias, uh, in the, uh, selecting patients, uh, participating in, uh, clinical trials. And also nowadays we are in the digital, uh, world, uh, the digital health

Sean: 00:17:53
is, uh, booming and is supporting, uh, patients, uh, in the real life and also in, uh, clinical trials. Then, um, how does the results communicate transparently to patients, uh, to those investigators and of course, not just to AI modelers, right? So these are the fundamental issues. Um, so, um, so for privacy, uh, I think, uh, the data, um, you're, you know, uh, we all know in healthcare, the patient data is, uh, De identified and anomalized, right? There's already material computer science algorithms systems for doing that.

Sean: 00:18:30
So we definitely to adopt this kind of technologies to make sure we do not leak personal private information. And also, when it goes to our models, when we select samples, then there's a lot of considerations about the human demographics. And how do we ensure that uh, The air models are inclusive, cover data points from different, uh, patient groups, uh, and separate populations. Uh, so that's a fundamental issue that we have to deal with when we do data gathering, uh, cleaning, uh, pre processing, and also when we do a model

Sean: 00:19:05
evaluation, uh, testing on, uh, you know, different, uh, patient groups. So these are the things that we have to, uh, go through, uh, definitely. And, um, uh, Uh, for clinical trials, uh, is similar things right now. How do we ensure that, um, you know, more patient populations, uh, different disease areas, different ethnic groups, different geographic regions, they could all participate to help test the safety and efficacy of the drugs because eventually the drugs when they approve, they will go to a much larger population. And, uh, so that's how, how

Sean: 00:19:39
do we, you know, create more. You know, robust, uh, results, uh, in the, even in the discovery and, uh, uh, clinical development, uh, process, we need a lot of, uh, involvement, uh, from different perspectives, uh, of the patients. Mm-Hmm. Uh, and, uh, and, and also, um, uh, we goes to the, uh, transparency, uh, communication. Uh, I think interestingly, uh, recently, uh, FDA has, uh, a guidance, uh, uh, in the, uh, about AI in medical devices. Uh, it, it was saying that,

Sean: 00:20:08
uh, the, uh, the results of. predictions from models, whether it's about the health outcomes and adverse events. So those things should be transparent across to all the stakeholders involved, which meaning including patients. Uh, investigators, uh, the AI modelers and, uh, abusing these people, uh, et cetera. So, um, uh, so I'm glad that, uh, you know, the field, uh, is starting to realize, uh, the importance, uh, that, uh, we should, uh, using AI to build, uh, abuse, uh, build trustworthy ai, we

Sean: 00:20:37
should, uh, you know, uh, be transparent and also engage the, uh, the end users. Mm-Hmm. , uh, in the process. So, so that we have more confidence about what AI. Uh, is working and what is not working currently and how can we, uh, find ways to make, uh, work even better? Yeah.

Hanh: 00:20:57
So true. Consumer focus. And include them in the, uh, upstream feedback. So that's awesome. So now you have received an SBIR awards for your data mining research. Uh, the small business innovation research program has been very instrumental in supporting cutting edge research in various fields. With over 43 billion in awards granted since its inception. Now, what lessons have you learned from your SBIR awarded data mining research,

Hanh: 00:21:28
um, that you think could be valuable for other researchers in the field?

Sean: 00:21:34
Yeah, I love this topic. So, um, as mentioned, um, When I was trans, trans, transitioning from a student to an investigator, I've gone through the whole process, how to apply, uh, the, uh, the grants, uh, from the government agencies. It's very much like, uh, the pitch to, uh, venture capitals, uh, for nowadays, uh, biotech companies, right? So, um, so to me, uh, there are some, uh, key elements that I learned. Uh, the first, uh, is to have the, uh, you know, the right teaming and then right technology, and then

Sean: 00:22:03
right, uh, um, objectives and tasks. And also eventually the right commercialization, uh, commercialization strategies. So, uh, so for example, um, you know, uh, there's a constant opportunities out there, uh, calling for proposals, uh, from government agencies, uh, from state, uh, agencies, from, uh, venture capitals, uh, et cetera. So, uh, then how do we, uh, grasp those opportunities? And the first, of course, uh, I mean, there should be a match of, uh, expertise and interest, right?

Sean: 00:22:33
And we select, uh, in the topics of example, a data mining in health care or data mining in social media analytics in psychology in human behaviors, right? We identify those solicitations, of course, and then once we have that, we should forming a strong, very strong team because, you know, the SBIR and any other grant applications, they need a very comprehensive views to cover the whole team. are the challenges, right? I mean, one or two teams or want to in the persons or individuals, they cannot do that.

Sean: 00:23:04
Definitely. So, uh, so what I learned is that, uh, yes, I realized I do have a lot of strength and I do have a lot of weaknesses, right? Uh, in real, in this kind of realization, and I, uh, I, I learned how to identify, you know, potential collaborators. You know, there's a lot of work to do networking, right? And when you show people, there's a grand opportunity to foreground. And if it happens to, um, you know, reside in a person's expertise, they're really very happy to participate, right?

Sean: 00:23:35
So because that's going to be a women's situation. So from this way, you need to identify What is missing, right? Because, um, you know, we talked about again, the technology stack, the data algorithm platform, you know, anything you're missing, you should be able to find someone who can back it up for you, right? Experts in the field. And once you got those kind of very strong points, elements you put together, it's time to cook a very nice proposal, then, uh, then, you know, we have to plug in the right technologies and in ground

Sean: 00:24:04
writing, you know, there's a way we have to, um, Um, describe the overarching goal on the list of business goals. And on each, um, a goal, we have to have a very detailed tasks, right? And then, you know, so basically, it's kind of a framework, right? We definitely need a framework. I mean, I mean, most founders, they wouldn't like those kind of ad hoc studies. They need a framework. Systematic frameworks that was a long term investment, meaning like you set up the infrastructure, you set up initial

Sean: 00:24:40
success of use cases and you keep adding new features, new use cases to that right? And also this is not sufficient because eventually we have to turn technology into impact, right? Then that needs a lot of, uh, you know, the business side development, the commercialization strategies, right? And, of course, for small companies, there's a lot of limited resources. Usually we have to partner with those bigger, larger organizations to do the commercialization. We show them our current technology and what is coming and see, you

Sean: 00:25:18
know, is that, you know, aligned with their, um You know, portfolios and the willing to, you know, looking to our technologies and collaborate and help and co develop the technologies, right? So, yeah, so it's quite important to find the commercialization strategies. So, given that all I think is quite exciting field, and I do encourage people to grasp those kind of opportunities, or at least think about that.

Hanh: 00:25:42
Mm hmm. Mm hmm. Well, I certainly have. Maybe we can talk about that afterwards. But, you know, with the advancement of AI, AI agents, and this whole ecosystem that you and I talk, How, what impact does that have in, um, putting together a team, uh, because a lot of the team members that we thought of was once important to have, can some of that be done or enhanced, um, through AI? What is your take on that?

Sean: 00:26:15
Yeah, so the teaming, um, So, yeah, you know, uh, I would say, uh, so, um, a I as a technology is definitely a fancy and then it relates to the business goals, whether we really want to get out of the business, right? That's business school. And also another quite important question. Part is a collaboration, right? As I just mentioned, you know how you put strength of people and forming a strong proposal, a strong program. So, um, I thought, uh, in in nowadays in AI technologies, they could not. I mean, those concepts are amazing.

Sean: 00:26:53
So, uh, I could find opportunity to collaborations and even through different ways of networking. I mean. There are so many ways that AI people, business people, um, by their industry or academic, uh, they can come together and, uh, you know, forming, uh, a team. And, uh, and, and, and again, I like the agility because I do believe, uh, the true scientists, they're coming from the, uh, the bottom up way, like from, uh, the, the front tier scientists who have been thinking this constantly and from experimentalists

Sean: 00:27:24
who've been doing the, uh, doing the experiments, generating data. Uh, things like that. So that maybe also need to formulate a cultural and also a welcoming environment to facilitate this kind of collaboration. So there's a lot of, uh, aspects. I mean, we can talk a lot. Yeah, it depends on the

Hanh: 00:27:47
people's angles, interests. Yeah, that's true. So AI and data science are poised to shape the future of health care and patient care. So now looking ahead, what excites you the most about the potential of AI and the data science to transform? Uh, patient care. And what do you think will be the biggest areas of impact?

Sean: 00:28:09
Yeah, yeah, this is a great question. So, you know, the ultimate goal is of health care and, uh, you know, farmer is to help with patients, right? So I think nowadays we are in a golden age. So, so for example, um, as mentioned, uh, like digital health can bring a lot of, um, medical devices. Uh, at home or your day to day work on your journey, right to help monitoring your health status and also the advance of biotechnologies, all kinds of omics that could potentially give us more

Sean: 00:28:41
insights into the reality of what is happening in different populations, different, different patient groups so that we can enable, uh, better Precise tailored medicines for different patients. So, um, and and also, uh, there's a very good timing, you know, especially through the pandemic. We all realize the health is quite important to our families to society. Then it's also good timing for the health care health educations through AI. There's a lot of learning. Uh, out there, uh, either on podcast or on the, uh, the article, social media

Sean: 00:29:22
and, uh, or, uh, seminars, people talking about, you know, all kinds of ways, how do we, uh, improve health because, uh, fundamentally, uh, I, you know, humans, uh, you know, we have, uh, our, uh, bodies that doing magic every day. But, uh, you know, there's so little, we really know what's really going on in the human body, in the tissues, in the, uh, cells and in DNAs, uh, and in cell division, uh, you know. you know, this kind of things. So, um, so I think it's quite important using this time using a I to You

Sean: 00:29:54
know, further increase the social awareness of this health issues. So then how human bodies works, what kind of food. Environment are good to our bodies, right? And so that way we can, you know, have the whole population society to be aware of this thing so that we can reduce the chance of having serious disease in the beginning, right? Um, and of course, things happens. Things are not working as expected. There could be gene mutations. There could be. You know, all kinds of ways aging, right?

Sean: 00:30:28
And these are all kinds of things could make us sick and sometimes getting into serious diseases. Then fortunately, then that's how farmer and health care are coming to help, right? So we develop those innovative medicines in case people we get into those very difficult medical situations. So there are still a lot of technology ways to do that. So I would say, think about this, uh, uh, now this, uh, AI, uh, booming age, uh, there's all kinds of ways, uh, that we could help with patients. And for patients themselves,

Sean: 00:31:02
they have all kinds of ways to, you know, engage the community, engage research, uh, researchers. And I'm glad, uh, all the, uh, the few that realize, uh, patient centric is, uh, uh, It's a huge, right, the way we all should be, uh, doing that. Yeah.

Hanh: 00:31:18
So true. There is no better time to, uh, with the available resources, um, guidance to have healthy aging, right? Yes. Um, there's abundance and what a great time to embrace the AI, um, to enhance your lives and longevity and with the tools and resources out there. Uh, even this conversation, for instance. So oh, it's great. Yeah. Yeah. Young researchers in AI and healthcare can benefit from the wisdom of

Hanh: 00:31:53
experienced professionals like yourself. Now, the World Economic Forum emphasized the importance of mentorship and knowledge sharing in fostering the next generation of AI talent. So now, what advice would you offer to young researchers looking to make an impact in the field of AI and healthcare?

Sean: 00:32:12
Yeah, this is my favorite topic, uh, because my, uh, I have my, uh, nephews, uh, entering college and were seeking advice from me. So I realize, uh, for nowadays, uh, young people, they, uh, kind of feel excited about AI, about other technologies, about, uh, healthcare. care, et cetera, but they are at the same time are kind of puzzled and and also I frequently I chat with graduate students before they graduate, I mean how to select your PhD topics and after you graduate, where do you go and you know how to jumpstart your career, so So I

Sean: 00:32:41
was recording on my own experiences, so I was thinking, uh, the key is to finding your passion and your strength, right? Uh, the passion, you may have a lot of interest, but what is your true passion? The only way to tell is to, you know, try that, right? Try that out. For example, for my PhD, uh, topics, I, I tried a bunch, uh, so many different areas. Then I finally, I thought, oh, okay. The biomedicine touched me better, so I chose it, right? So and then also find

Sean: 00:33:13
your strengths, right? What are you good at, right? If you're good at mathematics, you can think about, you know, this kind of AI modeling, algorithm development, if I model Into the, uh, uh, interactions. Then there's, um, you know, uh, you can, uh, think about those kind of, uh, the, the healthcare, like how to help, uh, be patient, uh, advocacy, uh, those, those kind of things. And you feel, uh, strong at electronic engineering how to build medical devices to help patients, right? So yeah, you need to find strengths

Sean: 00:33:43
and your, uh, uh, uh, passion and combine them to create, um. Your own portfolio. And also, um, I'd like to like to see, uh, to say, uh, be yourself and never give up, right?

Hanh: 00:33:59
So true. Overcoming setback is crucial, a crucial part of research process. So, um, from your experience, what strategies have you been most effective in overcoming setbacks and challenges in AI and data science research?

Sean: 00:34:15
Yeah, right. So there's many aspects, right? So, so there could be business aspects, business strategy, the market process, uh, and even, uh, I mean, so I, I, my interest is made on the technology and the science perspectives. So the, uh, the setbacks or the challenges I face is that, uh, you know, You know, again, back to the technologies stack and business applications. So, um, and many times we managed to build AI methods and tools put out there for, uh, users to, um, you know, test on that. So, um, you know, you get different

Sean: 00:34:50
responses from people and some people, they instantly Find benefits and would love to collaborate with you. So these are the great right? And many times you see, uh, neutral feedback or even negative feedback saying, Oh, this is, um, making wrong predictions, making wrong recommendations. I totally lost my confidence in those systems. I would rather. Go out using Google or using some other such tools to digest myself. So, um, so these are all boils down to that.

Sean: 00:35:23
Um, you know, yes, a is capable of the data. Is there? Uh, and the algorithm is there. It may or may not be sufficient enough. Then how do we, you know, build your initial successful use cases. So this need a lot of consideration. So again, my strategy is working with experts, right? You know, you know, before I before those intelligent agents can tell you anything. You know, where can you get the most trustworthy information? Of course, it's from people

Sean: 00:35:52
based hands on experience. Those experts, they're in the field, right? So if you collaborate with them and bring your AI technologies, you can, you can, you know, what is the role of the air? Is that going to be, you know, help me finding novel things that otherwise would never be always simply help me to prioritize things, right? So, so I would say Especially when we talk about a I lending in business cases. The key is that, um, also understanding the requirements and needs of the stakeholders.

Sean: 00:36:25
What are their real pain points? Maybe everybody thought, Oh, I is could have. That's true. But it's too general. What is the specifics? Right? Maybe once we understand those scenarios, having those kind of conversations, we It's like a diagnosis, right? We will know, okay, where's the pain points? Where AI can play? Where the existing data, existing models can, you know, help with?

Sean: 00:36:50
Then, what comes first?

Hanh: 00:36:51
Absolutely. It's not a silver bullet solution. I see it as, um, a GPS. It drives us. It uncovers insights. But we're the driver, right? We're the driver. Um, so, so we need to, I guess, provided quality data and make decisions from the insights that, that it gives us, but ultimately we provided input and we make decisions based from the AI insights.

Sean: 00:37:18
Yeah. Yeah. AI needs to be in the right hand.

Hanh: 00:37:22
Absolutely. And then in terms of, um, the use of AI, you know, this is what I always say to folks, you know, first and foremost, um, understand your, Your process. What are your pain points like you mentioned? Where do you see, um, the most nuisance or repetitive tasks? Or what is it that you want to remove from your day to day work so that you can get to the heart of your work? So then we'll focus on those pain points, you know, so because it's not a silver bullet for everything.

Sean: 00:37:54
So true.

Hanh: 00:37:55
Yeah. So now how do you cultivate a culture of innovation and collaboration with within your AI and data science teams?

Sean: 00:38:05
Yeah. So there's many different ways, right? There's some sometimes there's a bigger, uh, company level of initiatives. So these are kind of, uh, assigned, uh, You know, groups, they're working together, right? Then, of course, your social among those different groups. And among those groups, you can identify people that you have your best fit of interest and strength. And then you can pilot your studies, right? Build your own models.

Sean: 00:38:31
Show its significance, right? Show its accuracy. Show how it could be used in the business processes. So, uh, it's like making friends, right, in the open world, right? I mean, we all love to make friends, but you cannot make friends with everybody, right? In a research scenario, it's the same thing, right? Because the key is that everybody's goal is the same, is we need to have the win win situation, right? We have, you know, one plus one

Sean: 00:38:54
is much greater than two, right? I mean, everybody's looking for that. So, if we set up those kind of tone, and the people know that, and the people can bring their own ones into the table. Then it's very good timing to cook a lot of fancy things, fancy foods, and that will be much greater than two. So I would say, uh, be, you know, be open, be ready and, uh, you know, be proactive to find those opportunities and making our connections. Yeah.

Hanh: 00:39:24
Very true. So what role, uh, do strategic partnerships play in advancing AI and data solutions in health care and how? How can organizations effectively foster these collaborations?

Sean: 00:39:37
Yeah, I think that's so a big topic and so such an important one. Um, I can only share my personal views. So, so the strategy, usually, to me is aligned with the business development. I mean, if a farmer, every farmer, they have their own portfolios of disease areas, their own molecules, their own clinical trials, et cetera. Right, then, then this usually coming from, you know, the business, um, You know, assessment. What is the short term near term goals of the companies? And then the technology is

Sean: 00:40:12
kind of, um, go afterwards. And sometimes I see technology can go ahead. You don't have, like people say, uh, AI driven, AI based biotech companies. They purely work on AI predict targets. molecules. That's so true. But that's kind of can be afford as this environment by those very small companies, right? Because bigger companies that have many more resources, a lot of expertise, right? We never want to, you know, bypass those kind of resource and expertise to pursue

Sean: 00:40:46
something in, you know, uncertain way. So, um, So So how do we form the strategy? I think there's different levels, right? There's a company level. And even like in the farmer field, I saw a lot of alliances. Companies, they come together, share their, uh, you know, uh, data in an uncompetitive way, like, for example, if they have this kind of early discovery molecules about this data, what works and what not works, there's a concessions like by some farmers. They do share, of course, those kind of early discovery data.

Sean: 00:41:21
Uh, and also, you know, government level, of course, and FDA is carefully watching how AI field is, um, progressing and constantly updating their guidance. Uh, and, uh, and when it goes, uh, yeah, then we go back to the inside the organizations are different teams than the strategies. So again, I think, uh, we are all limited by resources, cost, time, budget, et cetera. But now, how do we. Uh, create a winning business cases. Then again, I think this depends on the people, right?

Sean: 00:41:53
I mean, we have to bring together the right people, the right expertise addressing the right business questions. So that way we can get a handle. And, uh, we can show, uh, our early success, broadcast, uh, broadcast success so that way we can attract more people to join the journey and eventually, you know, can improve the, uh, efficiency for the whole industry. And then, you know, there's definitely, uh, we are in a time of revolution. I can see. Yeah.

Hanh: 00:42:22
Absolutely. And many, um, many folks are still. Going through the paradigm shift, it's necessary, but, um, the time is now just don't wait too long. Right? Because I think consumers are getting really smart. Um, they're going to expect AI integration, whether it's in their customer service, information provided to them, um, so that they're armed and equipped to make decisions, whether or not to use your service or products. So it's really, um, empowering

Hanh: 00:42:51
the consumers, the patients. And, um, I think that's key. Um, with the advancement of AI, I think consumers are going to expect that. Yes. So now balancing innovation with patient safety and privacy is a crucial consideration in healthcare AI. So as AI continues to advance in healthcare, how do we strike the right balance between driving innovation And, um, ensuring patient safety and privacy?

Sean: 00:43:27
Yeah, excellent topic. So, um, you know, ePharma, right? I mean, we all know the goal of the clinical trials is to test the efficiency, uh, sorry, the efficacy and safety in patients, right? So, um, I would say, uh, there's many, uh, different angles. Right. Uh, and, um, and nowadays, uh, I think the, the, the, these are all coming from the unmet medical needs. And sometimes there's a new, no treatment for a certain disease. And sometimes there's, uh.

Sean: 00:44:01
There are marketed drugs for indications disease, but there's a drug resistance issues, um, toxicity issues that we have to constantly looking for new targets, new molecules, uh, to, you know, address those, uh, medical needs, um, and, and during this, uh, process, you know, um, so there's a lot of, uh, considerations, I would say, um, Uh, it again. I would like to introduce the emphasize the importance of patient centric way, right? Because, uh, nowadays when we previously I mean, it's purely like the develop the R. N.

Sean: 00:44:44
D. Separate from the patients. I mean, when it gets to the patients, it's the drugs already on the market and prescribed by the doctors. And nowadays, um, innovative companies, they are engaged. The patient voices in the very beginning, like, uh, Because one target, one molecule could work for some population, but for specific patient populations with different biomarkers, different mutations, their, you know, their safety profiles and ethics profiles will differ. Now, how do we engage those

Sean: 00:45:12
patient voices, doctors voices into the very beginning? Uh, that's, uh, I think that's a field, uh, is, uh, uh, like, uh, Working on and, um, and of course, uh, we talk about the safety issues that's unavoidable, right? And because medications, um, they get into the body and, uh, usually they, I mean, many, most of the time they have, uh, onto the tachys, those drugs, they bind to these, uh, uh, the genes, uh, to achieve therapeutic, uh, proteins to achieve, uh, therapeutic effects, but sometimes there could be. Off topic issues.

Sean: 00:45:51
And nowadays, uh, when the society is getting, uh, aging, there is a good, a lot of, uh, drug combination issues. And, uh, some drug combinations, they were not, not, never tested before, and they could have unexpected adverse events. So this again, uh, relates to, uh, regulation and, uh, you know, reporting systems, uh, ensure that we know those adverse events and, uh, think of ways how to prevent that. Yeah.

Hanh: 00:46:15
Mm hmm. Very true. So, now, um, natural language processing has numerous potential, uh, applications in the health care. So, a recent study published in the, uh, Journal of Medicine, Medical Informatics Association, it highlighted the promise of NLP in extracting valuable insights from unstructured medical data. So, what do you believe is the most promising application of NLP in health care? And why?

Sean: 00:46:44
Oh, yeah. I love this topic because I have been working on NLP for so long. So, um, let me give some brief examples, right? So in the pharmaceutical R& D, there's a lot, definitely a lot of contextual information document, uh, uh, you know, in early discovery in clinical trials and also in real world evidence. So, uh, in early discovery, as mentioned, uh, uh, you know, researchers constantly need to return to the, turn to the, uh, literature to find, uh, novel information to, uh, help, uh,

Sean: 00:47:12
facilitate their hypothesis generation. So, uh, NLP, you know, or previously called text mining technologies, then can help you extract, uh, key information about gene disease, uh, molecules and summarize that for you. And also could potentially, uh, extract the clinical trial design and results, uh, results, uh, sections. Uh, so that way, um, clinical teams are, are, are, are, Uh, early, uh, discovery teams, they can digest this information, uh, much quickly than they manually curate this information. And a clinical trial documents is

Sean: 00:47:48
another, uh, key, um, uh, field where, uh, NLP can be applied. Uh, so, uh, a lot of, uh, protocols, a lot of, uh, clinical trial design documents, results documents, and how do we, you know, using, uh, NL, NLP, LLMs to translate that into structured information. You know, you know, very nicely, uh, format, uh, tables, um, you know, uh, and reports. So, uh, definitely, uh, there's a lot of work going on and some proven already, uh, some products are already in operation. People using that to generate those kind

Sean: 00:48:22
of, uh, uh, extraction results, reports. And when it goes to the, um, real world evidence, nowadays we use, uh, electronic health records in EHRs. There's a lot of unstructured information as well. Uh, like, uh, uh, clinical notes, uh, admission notes, discharge summaries, et cetera. So NLP can potentially use this, um, you know, uh, work on those kind of, uh, information, uh, to help predict the, uh, uh, future outcomes. And also those information could be digested and inform,

Sean: 00:48:54
uh, the, uh, clinical trials. So, uh, so these are the classical, um, applications that's already going on. Uh, and, uh, even before AI was. It's called AI nowadays, like, uh, like in 2013, 2014 times, there's a lot already kind of text mining technologies enable us to do that. Now the difference is that with deep learning and NLP, we can get even better accuracy and more flexibility, uh, because, um, you can, you can. You know, using, uh, this kind of ities, using this kind, uh, this kind of things like, uh, prompting, uh, engineering,

Sean: 00:49:26
uh, using, uh, this racks, uh, with uh mm-hmm, retrieval, augmented, uh, uh, uh, uh, the, uh, uh, generation, right? You know, those kind technologies have you give you the flexibility to, you know, interactive, uh, inactively working with models. So those are the differences. By essence, people already realize that, uh, you know, NLP could help with, uh, productivity. And also, I'd like to highlight NLP is essentially a sequence of models, right? Now we use, uh, you know, the word, uh, the, you know, documents of words and,

Sean: 00:50:02
you know, a sequence of words, right? That gives a meaning, a semantic meaning of the document. But now, uh, if we adopt. The sequence models concept, then there's a lot of things actually can be modeled as a sequence, for example, for example, image processing, right, actually, transformers can also be applied on image data. So what people do is that they can do a segmentations of the images, uh, image into, um, A list of patches and those patches is again sequential data that can input into your, uh, transformers

Sean: 00:50:33
so that you can pre training your model and finding your model, et cetera. Uh, and also, uh, excitingly, um, nowadays, we also using those kind of NLP, uh, foundation models in biology, because, um, think about that, uh, you know, just like a words. The, you know, they play together, they interact with each other to become a document. Inside our bodies, you know, the genes, the interact with each other to determine the cell states and cell, uh, cell communicated with each other to determine biology, right?

Sean: 00:51:10
So those things, if you think about that, they could also be potentially, um, transformed into, um, uh, transformer, uh, kind of pre training, um, Uh, paradigm, right? So, so, so in this sense, uh, NLP no longer, uh, because just one text model, it's become a multi model and, uh, it's, uh, goes far beyond text. Yeah.

Hanh: 00:51:34
Well, thank you. Thank you so much. So we're at the tail end of our conversation. Would you like to add anything before I close up?

Sean: 00:51:42
Oh, thank you so much for the opportunity. Um, so, uh, I would say we are in an exciting, uh, uh, area, right? With all the data technology, business applications and medical needs out there. So I think the key is to have the right execution strategy. If we can have the right strategy to put together the data, algorithms, platforms, business use cases, uh, without. Be able to significantly enhance and transform the pharmaceutical R and D and also many, many other, um, uh, societal issues.

Sean: 00:52:21
So I think, uh, and the good thing is that AI now becomes ubiquitous and everybody use GPT or things like that, right? And so, um, You know, these contrast, uh, previous years people talking about keep asking, okay, how to democratize ai. Now, people don't ask this question. The AI is already in your hands, right? So now I do encourage, uh, people to grasp that whether you are, um, uh, tech, uh, technicians or, or technologists or you biologists or you have patients, you will find, uh, the benefits of the ai, uh, uh, from here and there sooner later. So I encourage empathy,

Sean: 00:52:57
embrace this kind of culture. Yeah. Thank you so much.

Hanh: 00:53:01
Well, Dr. Liu, thank you so much for joining us today and sharing your invaluable expertise and AI and data science in pharmaceuticals. Your work is truly revolutionizing drug discovery and development. Bringing hope to countless patients worldwide. And to the folks who are listening or watching, we hope that you found this episode as inspiring and informative as we did. Dr. Liu's dedication to advancing

Hanh: 00:53:28
health care through cutting edge technologies It's a testament to the power of innovation and collaboration. So if you enjoyed this conversation, please like, subscribe, or leave a comment below. We'd love to hear your thoughts. So join us next time as we continue to explore how AI is transforming the lives of the 50 plus demographic. So until then, stay curious, stay engaged. And remember that the future of healthcare is in our hands. So thank you so much for tuning in to the AI50 podcast.

Hanh: 00:54:05
Take care.