Town Hall: AI in Pharmacy

[Kathy Giacomini] Okay, so hi, I'm Kathy Giacomini, and I am going to host the first part of the town hall today. So welcome, everyone. And before I begin, I would like to comment on the world today: And I would like to have just a moment of reflection. As a health sciences school, our primary concern is to do whatever we can to reduce suffering in the world. And many of us are still struggling to comprehend the atrocities that continue to unfold in Israel and Gaza. Personally, as a dean, I have reached out to a former pharmacist scientist student of mine who lives in Israel, he is going to try to put me in touch with deans of schools of pharmacy in Israel and Palestine, and perhaps there will be ways that our school can help in that area. So that's our moment of reflection.

Today, what I'll do is I'll give an update, and then we're going to have a lively discussion on artificial intelligence, and it will be a panel discussion presented by a number of people that Eric Davila will be introducing and moderating.

So first of all, I just want to mention that today, or today this week, we're in National Pharmacy Week, and this month is American Pharmacists Month. So it's a very timely moment, if you will, to have a town hall in a school of pharmacy.

So let me comment on who has retired this past year. We've had four retirements, from faculty and staff: Lucia Piriano, Marilyn Stebbins, Kris Casler, and Deanna Kroetz have all said goodbye to us, although some of them are very much in touch with us. We have new faculty appointments: Willow Coyote-Maestas just started, I believe this fall or past summer. Leslie Floren has started as a faculty member in the Department of Clinical Pharmacy, and Sook Wah Yee has become a faculty member in the Department of Bioengineering and Therapeutic Sciences. So welcome to them and, farewell to our recent retirees.

I also want to recognize many faculty, staff, and students have won various awards, and I may miss some of you here, but I'm highlighting a few of you have won awards: Jaime Fraser became the new chair of the Department of Bioengineering and Therapeutic Sciences, and Nadav Ahituv was appointed the director of the Institute for Human Genetics, Nicole Flowers was awarded the 2023 Chancellor Award for Exceptional University Service, Jim Wells received an Honorary Doctor of Science from the University of Chicago, Nevin Krogan received the French Legion of Honor Award and the Discovery|Innovation Health Prize from Research!America, and Katherine Gruenberg and Stephanie Hsia were named 2023 American Association of Colleges of Pharmacy Emerging Teaching Scholars.

So we've had a range of awards. If this were live, I'd say everybody give them all a hand, but it's not, so let's just recognize them all. I also want to recognize Sharon Youmans and Jaysón Davidson. They are the 2023 Chancellor Award winners for the Dr. Martin Luther King, Jr. Leadership Award. So this is a huge honor, and this is just announced, so we should recognize both of them for this absolutely amazing award for both of them. Jaysón is in the Pharmaceutical Sciences and Pharmacogenomics graduate program, and we all know our own Sharon Youmans on our faculty.

I also want to recognize students—three—first of all, one set of students, a big team, won the Value of Industry Pharmacists Case Competition. I love it—they won first place! So I'm not going to read all their names, but it's a great, excellent team. I know some of these, they're very engaged in pharmacists in research and in the pharmaceutical industry. We also had three students were named champions in the American College of Clinical Pharmacy. It's called the Clinical Research Challenge, and those three students are Ryan Sue, Pollyanna Leung, and April Zhou, so our students are doing absolutely amazing things around the country.

Now let's move to our next...Ryan Hernandez is organizing a PhD Student Diversity Fellowship Symposium. Last year at a town hall, I announced that we had just received a five-year award from Genentech to award two fellowships for graduate students in any one of our graduate programs who are underrepresented minorities. And last year, those awards were given to Brenda Melano and Sebastián Cruz-González. So at this symposium on October 30, that Ryan has organized, both of these two individuals will describe their experience, which included a rotation at Genentech, and I'm hearing from Ryan that they loved it. So please go and attend this and listen to their amazing experience at Genentech and as a fellow.

Other updates include...we continue—and this almost sounds like a little bit boring—but we continue to be the number one school of pharmacy recipient of NIH funding in fiscal year 2022. And that was for the 43rd consecutive year! I don't think there's anything like this for any other health science school. And one of the grants that we have that we renewed this year for a five-year renewal is our UCSF-Stanford Center of Excellence in Regulatory Science and Innovation. And that was renewed for five years for up to $10 million per year. And I'll say a word about the summit that they that they sponsor each year. But before I do, I wanted to say that one of my goals as dean was to celebrate our entrepreneurship. And so we now have a website. This was Eric Davila, working with Adam Renslo. They have a website now, they created a website in our school, which really showcases 38 companies founded by our faculty, maybe students, and staff. So this is great. We're going to do more of this with time.

A moment on our 2024 CERSI Innovations in Regulatory Science Summit: This takes place and has taken place for the last few years on the January right before the JP Morgan Conference, which takes place in San Francisco. And so it's Sunday, January 7, it starts at 9 a.m. And it is a fabulous conference. It brings to UCSF the current FDA commissioner Rob Califf (he is very likely to come again this year and never tells us till the last minute) along with former FDA commissioners. We even have an FDA chief's chat, which is just great, as they talk about issues that the FDA is facing. We also have a panel discussion on why is medical misinformation killing us. And we're also having an FDA center directors—all three center directors from the Center for Drug Evaluation Research, Biologic Evaluation, Research and CDRH (Center for Devices and Radiological Health)—all three of them will be presenting in this forum. So that will be great. We're also having debates this year: Should advisory committees always vote? and Is direct-to-consumer advertising good for patients? So this is going to be a very exciting day and I encourage everyone to attend. It will be hybrid, but of course we'd like to see everyone in person. It's going to be over at Robertson Hall and our community center at Mission Bay.

The master's degree program, which you'll be hearing about today [was] started and developed by Joanne Chun, along with scientists at Genentech and our own faculty, was approved by the UCSF Academic Senate about a month ago. It's now at system-wide for approval. So fingers crossed, it's called Artificial Intelligence Computational Drug Discovery and Development. Joanne is hoping that it will be approved and we can begin recruiting our first class, which will begin in the Fall of 2024. We're also actively involved in developing a second master's degree program—again, this will be available to our PharmD students. This one will be in Regulatory Science and Health Equity. So we're working on that this year. We hope to get that through Graduate Council (GRAD) at UCSF this year. I now want to turn it over to our exciting panel discussion on artificial intelligence, which will be moderated by Eric Davila.

[Eric Davila] All right. Good afternoon, everyone. Thank you for joining. Generative artificial intelligence and machine learning has exploded in the popular consciousness in the last year, and many are working on and wanting to know what that holds for the School of Pharmacy, for UCSF, for science, and the world in general. So first up to speak with us today is Ida Sim, who is a primary care physician, informatics researcher and entrepreneur, and has been leading some of the efforts around AI and what it holds for UCSF. Here at UCSF, Ida.

[Ida Sims] Great, thank you so much, and thanks for this opportunity to speak to the School of Pharmacy. It's wonderful. So I will just speak to AI at UCSF, and not the world and all the implications, because that is really so much. Just to set the research and funding context: Great that the School of Pharmacy [received] the 43rd year in a row of being first and NIH funding. We want to keep it that way across the whole campus. And as you know, discovery and clinical research grants are increasingly incorporating AI and ML, if you did a really simple keyword search on machine learning and titles...growth of 450% in NIH grants, and, you know, way above that for 2023 already. Of course, major grants from ARPA-H (American Rescue Plan) HNSF, and NIH, and so forth. Also getting into AI and ML.

Implementation science grants: It's not just machine learning and AI, I'll just say here. There's analytic AI, which is what I think the National Academies are calling the "non-generative AI." So analytic AI is the machine learning that you're probably more familiar with, you know, predictive analytics, there's planning, there's a whole set of traditional AI that comes under sort of "analytic AI." And then generative AI—the newer stuff—that transformer-based, large-language models where the output isn't a number or a classification, the output is natural language. It could be a picture, it could be music, it could be code. So generative AI is really quite different. And it seems like the terms are now generative versus analytic AI.

But even beyond AI, there's technology also...implementation science grants increasingly require integration of digital health and digital support into the health system. If you did have an AI-based decision support system, how do you get it to the patients? How do you get it to the clinicians, to the pharmacist at the frontline? And more and more grants need that, too. So UCSF really needs to build this first-class infrastructure to attract and retain talent, and to maintain our research, leadership.

So in my role as Chief Research Informatics Officer—I've been in this role for about six months now—this is the broad vision for AI, which uses computational technologies, and analytic and generative AI to advance health worldwide, both in the discovery sphere and in the clinical sphere, in anticipating what and how research is to be conducted, and do it in a in a fair, equitable, compassionate and trustworthy way.

So it's helpful, I think, to define this sort of white space of...I think we're all somewhat familiar with what IT is, but what is this space of AI and digital health and technologies and data-sharing, and, data federation, and privacy, and all that kind of stuff. [We've] come up with this term Knowledge Computing: scientists designing, developing, evaluating, and applying specialized algorithms, AI, machine learning, and hyperscale scientific data. My colleagues in discovery sciences say that the scale of data and discovery, it is actually larger than what we typically deal with in clinical care and EHRs (Electronic Health Records) for example, and infrastructures to support that kind of computing.

And what's the purpose of this? It's to augment human cognition. Is it to do discovery? Yes. Is it to do clinical care and education? Yes, but really, this kind of computing is most powerful if it's paired with our own thinking and our own cognition. Phrasing it as "augmenting human cognition" puts a different spin on what it is that we're doing, and how we evaluate the output. The output isn't just what the system does, it's what the system in addition to the human does.

So Knowledge Computing is really quite different. Many of you are probably familiar with a stack way of thinking about it, where you've got hardware hosting networking security on the bottom. I would call that the responsibility of Enterprise IT. And then I'll talk about Research IT, and you're familiar with that as well...high-performance computing, BRIDGE, HIPAC, some of these systems you may be familiar with. If you do clinical research, clinical trials on OnCore, iRis, you're probably familiar with this if you do human subjects research.

And then on top of this is Knowledge Computing, where we're really getting more and more into augmented intelligence. And we do all work together, but I think it's helpful to think of Enterprise IT as being under the remit of CIO Joe Bengfort. The Research IT scope is under the remit of Mandy Terrill, who is our Associate Chief Information Officer for Research. And then me, Chief Research Informatics Officer is at Knowledge Computing. All components of the technology stack are absolutely essential. So you'll hear us all talking together and planning together. My particular objective is really to drive as much as possible toward the top of the stack.

And you say, "Well, what's at the top of this stack?" Broadly speaking, the idea might be to go from molecules to communities. Can we actually reason across the whole spectrum of what we do with our discovery side of campus? And then with the health side, and also the clinical research side as well, we tend to work very much in silos, even within discovery. We work in silos—proteomics, versus genomics, versus, you know, all sorts of silos—and with large language models and new technologies, there's a potential for bridging those in ways that we haven't been able to before.

As a clinician, I would like to see discovery come more into our clinical world, certainly with the pharmacogenomics deployment in our clinical systems, we're starting to see that, but let's close the loop. If we learn something, can we drive it back into discovery? We're not set up in this way at UCSF right now. And really, what's changed, what Sam Hawgood, has charged us to do, is to identify leapfrog opportunities. What can we do with these new technologies that we weren't able to before? So this is sort of a big picture. And obviously, you say, Well, that's very aspirational. Indeed, it is.

What are some more concrete things? So first of all, generative AI has come to UCSF. Some of you have heard about it. This is a tool called Versa, that's at the bottom left here. This is Microsoft's Azure implementation of OpenAI. So this is the OpenAI model, we have it in our own secure Azure environment, so that we can actually put in secure data. Even your patient data from UCSF, for example, can go in there. No one else is going to get it. You don't stick it in there and, you know, Google is not going to get at it, Microsoft's not gonna get at it, OpenAI is not gonna get at it. It's been rolled out to a limited set of users, we are incrementally opening that up as we learn more about capacity, and so forth. So do look out for it over the coming months. You can also sign up on the waiting list at http://ai.ucsf.edu. On the bottom right you see some training and background information on AI at UCSF, and just general background as well. There's going to be more and more resources, so this is one place for you to go as a good landing spot to start with what we have on offer and UCSF right now.

For those of you who are interested in deploying AI into the clinical environment, Sara Murray, who's the newly named Chief Health AI Officer, runs the HIPAC, the AI platform for APeX. It is a secure, cloud-based computing infrastructure. You can deploy your AI algorithms in it, it does, obviously, get data from APeX in real time, and also from our Chronicles and retrospective data as well. So do email [email protected] for any questions about that.

I would say that, of course, there's a concern about safety and ethics and fairness of these algorithms that are being deployed in the clinical world, and I would say similar concerns if we deploy some kind of AI for education, for example. My office along with Sara's office, we're starting up an algorithm evaluation and monitoring workgroup. And there's a tremendous amount of novel methodology that's needed in establishing the effectiveness of AI and validating it, but then also making sure that the performance of the system doesn't drift into unfairness, inaccuracies, and so forth. That's something that's pretty new for us to be doing. It has infrastructure implications, and methodological implications, and scientific implications for how we design our systems, how we design, the evaluation, and so forth. Then, of course, ethical issues underlying all of it as well. So this is going to be a big, big stream of work. We are just starting it. Those of you who do high performance computing, are perhaps familiar with the work that Mandy has been doing in this in this space, in this area

We're moving towards a high-performance computing core, which offers several different environments for high-performance computing. Certainly, Wynton is an important environment, and that's the current UCSF environment, which quite a number of faculty are on. Faculty also have their own high-performance computing clusters, and we're trying to bring things together to offer centralized support in a way that makes sense for faculty. And so right now we are looking at a new cluster, which includes data storage, and both CPUs, and also these much more powerful GPUs that will allow us to do the kind of computing that we want to do, especially with large-language models requiring just tremendous amounts of compute power.

This is something that's very, very high priority for the campus and for Chancellor Hawgood. We're looking at places to host this physically, and it comes down to power and cooling and rack space that we can actually, you know, install these GPUs—if we can even buy them—because there's a big shortage of them. So, very high priority for the campus, we know that that is the engine upon which a lot of the research now and in the future is going to be powered. And of course, we're looking at governance and oversight of the core as well. So a lot more to say there, and I'm sure at some point you can hear from Mandy on that.

So where I'm thinking about what I'm doing in the CRIO [workstream], there's the Applied Ethics—I mentioned that in terms of developing principles—and it's not just that there's so many AI principles out there, it's really how do we implement them in risk-based frameworks that support what we do as researchers, and support our patients in our community with fairness. Training is going to be critically important, evaluation-monitoring, I've mentioned. I'm gonna skip to community, we're thinking of setting up communities of practice where those who have ideas and those who have AI expertise can come together so that we can address the kinds of problems that really do need to be solved. And of course, websites, communications, and so forth.

On the Architecture front, I have been in discussions with several of you on this on the Zoom, about ways to think about how we architect data and knowledge and compute at UCSF so we can do, science that has not ever been possible before. I'm not going to say more about that, but that is something that is of high interest. We're going to be going for a National Science Foundation AI Center grant, for example, and working closely with the health system on research platforms that allow us to do much more agile clinical research at the point of care.

So I'm gonna stop there, and just say stay tuned, there is going to be a wider rollout of Versa, which is the ChatGPT. There are two interfaces, one is ChatGPT, where it's just like you're familiar with on the browser. And then there's also an API access, seeking a programmatic access to the OpenAI foundation model. We will be bringing on LLaMA 2 as well, for those of you are interested, this is all within Microsoft Azure. So watch for that as it rolls out. There will be guidance and training on AI best practice. We'll be developing more around applied ethics. And then, concretely, coming up, there will be a series of lunch-and-learns over the fall and winter on generative AI and getting started with AI.

We're also in very close touch with the IRB (Institutional Review Board) about what new principles or applications are being needed for IRB, for data-sharing, for safety, and so forth. So all of those are in process. This is something that is really very new, but very exciting, and critical for the institution to build on going forward. So I'm going to stop there and take any questions or comments.

[Eric Davila] All right, we've got one question so far. And obviously, if we get more later, I should have some time at the end. But you said—I think you may have touched on this—you said that the output, it's not only relying on the system, but the phrasing and the prompt for the AI. It seems like a pitfall, potentially, when replicating studies or doing testing. So can you give any insight about how human error could be harmful or undermine studies?

[Ida Sims] Oh, wow. That's a broad question. [Human error] is in all kinds of research. I think that there are different types of risks that we have and different sources of error. But fundamentally, there's going to be bias, there's going to be inaccuracy, there's going to be uncertainty. And there are methodologic ways to think about that. And again, I don't want us to focus too much just on large language models. There's a broader set of technologies including causal inference and many other methods that allow us to think about how we make inferences, how we give decision support, and how we do it in a careful way. My comment about the system is really that the output isn't just "Does the AI make the right recommendation?"

Really what we need to evaluate [is] does the AI make a recommendation, and then how do humans—is it the clinical pharmacist or the physician or even the patient—how do they respond to that, that decision support and what actually happens? It's a combination of the AI system and the human that generates the actual impact in the world. And so I think we need to be thinking broadly about the inputs to the impact that we want, and not fixate over much on the computational system, particularly because with generative AI, the output is often very human-like, right? It's language and how that intersects with people, and how people then think, I think is really a new area and introduces a lot of new risks, and a lot of new biases, that we need to think about both individually as researchers. But I think there's some institutional responsibility as well. And I think we want as UCSF to be leaders in thinking about how these technologies can be deployed, and how to keep our populations or our patients or communities, both safe, and also as healthy as possible.

[Eric Davila] I have one from Dean Giacomini, who wants to know if there's similar people with your role on the other campuses? And do you work with them, and in what context?

[Ida Sims] Other academic medical centers? Indeed, there are. And they are often, I would say, more on the Research IT front, they're very, very IT, I am explicitly not IT, and that's kind of why I defined this Knowledge Computing term. Because it's really looking at the newest technology and applying it across the board to the way we do science. It touches very much with Research IT, but Research IT is more enabling and supportive in a sense, but Knowledge Computing is very strategic in terms of the kinds of science that we want to do.

[Eric Davila] All right. Ida thank you so much. I'm gonna move on to the school portion, the next portion here of the program. But, time allowing, we will come back and answer some more of these questions. And if not, we will try to follow up directly or in chat, or at the next town hall. So now we're going to find out a little bit more in a series of lightning talks from some various leaders throughout the school, really just the tip of the iceberg of people working on AI in the School of Pharmacy, starting with Joanne Chun, who is Academic Coordinator and leading the charge on a new master's program.

[Joanne Chun] Thanks, Eric. So today I'm really excited to talk about our new program in the School of Pharmacy. This is the master's in Artificial Intelligence and Computational Drug Discovery and Development. And, as Kathy mentioned, the program has recently been approved by our Academic Senate, and it is on track to launch in the Fall quarter of 2024.

So, the landscape of drug discovery and development is rapidly evolving and becoming increasingly computational. This is driven by advancements in technology and the increased use of computational methods in the field. And as the field leans more towards data-driven approaches, we're seeing an increasing demand for scientists who are trained in artificial intelligence, machine learning, statistical techniques, pharmacometrics, and other data-driven methods.

Next, and while job opportunities in drug discovery and development continue to rise, especially in the realm of computer and information research, there's a noticeable gap in academic training programs that are available, and our program in Artificial Intelligence and Computational Drug Discovery and Development will address this growing demand for scientists trained in computational methods across all areas of drug discovery and development. So our program will be open to applicants with a bachelor's degree. But students with a PharmD will be especially well-suited because they already possess a foundational knowledge in pharmacology and human disease. And this program will create new opportunities for these graduates.

So this will be a one-and-a-half-year program consisting of a total of five quarters. Students will spend the first three quarters taking didactic courses, and then they'll spend the following two quarters working full time on a capstone project, either in industry or in academia.

Next, our curriculum offers cutting-edge courses, and ensures students are equipped with the latest knowledge and skills in computational drug discovery and development. And here we show some curriculum highlights. Our course CDD 201 focuses on techniques used in drug discovery; 202 teaches programming skills in Python; 203 delves into artificial intelligence and machine learning; 204 focuses on regulations related to drug development, and introduces relevant software platforms; 205 teaches applications of pharmacometrics, systems pharmacology, and pharmacogenomics; 206 is about big data mining and analysis, specifically in the context of real world data and evidence; and 207 focuses on modeling techniques used in clinical pharmacology; and 223 highlights emerging technologies.

So after completing these first-year courses, our students will then take a deep dive and explore areas of interest while working on a capstone project for two quarters. And importantly, we've established collaborations with industry partners, and we have leading experts who will provide guidance and feedback to the program, as well as offer capstone project opportunities to our students.

Next, and our graduates from the program will have the training to apply computational tools and methods across all areas of the drug discovery and development pipeline, so from drug discovery, drug development, bioinformatics pharmacovigilance, to translational medicine, and they will be ready to emerge as leaders in the field spanning academia, industry, and startups. So this concludes my brief overview of the new program. Thank you.

[Eric Davila] Thank you, Joanne. Next up, we have Conan MacDougall, who is co vice dean for PharmD education. He manages the PharmD program along with Igor Mitrovic, and he has been working on the question of what AI means for education, I believe.

[Conan MacDougall] Something like that, unlike Joanne my focus will be primarily on our professional curriculum, as well as our faculty and staff that support that. So we have an education workgroup. Members are listed there. We have representation across both our faculty, and representation from pharmacists at the medical center, and then we have student representation as well.

We are sort of thinking about this in terms of three phases, And the initial phase that sort of happened when everybody realized that you could use ChatGPT to cheat on all your assignments was to think about what is the allowed use of artificial intelligence in our educational program? And so we've been developing a policy on AI use and assessments. Importantly, given the big role that AI will likely play in the future, our policy is designed to have flexibility to make sure that faculty can deploy AI and assignments when they're working with students.

It's really about developing a shared understanding and knowledge about ways in which it's appropriate. So this proposal's been presented to students, the directors of our different themes in our PharmD curriculum, the dean's leadership group, and is with our curriculum and educational policy committee, and we hope to have approval soon.

We also need to develop general understanding of the basics of AI, as well as an understanding of what this policy—when it's approved—means for our educational endeavor. And so we're working on presentations that will come for students as well as our faculty and staff.

But we really hope to transition quickly away from just sort of use/don't use/when can I use this, to thinking about what the productive use of AI in education can be. From a student development perspective, we have some of these core concepts in artificial intelligence teaching sessions scheduled for our P1 and P2 students, at the end of the year, and the beginning of the year. This is really a couple of hours for them to just understand, really, what are the basics of AI from almost sort of a layman's perspective, and what are the some of the ways in which they can interact with it.

We also have, thanks to Joanne's leadership, a set of inquiry immersion courses, where students learn more about a topic in depth, and one of those mini courses will be around artificial intelligence. And so a small group of our P1 students, about 12 students or so, will learn about AI to a greater degree of depth.

All of these materials, as we develop them, we will then use and repurpose to educate different audiences. So we have preceptors at the medical center that we're also going to talk about AI with—both basics of AI, and then how they can use it in education. That's going to happen next month. Later in next year, we'll have a faculty retreat, and hopefully we'll be able to do some education, to the faculty and staff then. The American Association of Colleges of Pharmacy has an institute—two days for the virtual workshop—where they'll be talking about deployment of AI and pharmacy education. If you're a faculty member, and this is something of keen interest to you, reach out to me and we can try and connect and see if that's something you could participate in.

We also need to develop our curricular outcomes around this. What is it that we want our students to learn, and where in the curriculum should they learn it. And then we hope to be able to develop a resource center in skills-based training for our faculty and staff so that they can explore more on their own.

And then we're looking forward to the part where we integrate AI throughout the educational enterprise. There are some assessment and simulation products, that are...both early versions are available, and then at some point, more sophisticated versions will be available, to do things like create your student exams for you. And so the questions will be, in what ways should we use these tools and take advantage of them? And in what ways should should we sort of use these things on a faculty side?

And then thinking about, again, what does it look like when AI becomes sort of part of the enterprise? Right? Pretty soon AI is...it's not gonna be like, "Well, I'm using AI," right? Like I'm entering stuff into ChatGPT. Because it's going to be a part of Gmail, it's going to be a part of Microsoft Office suites. And so what does it look like when AI is integrated across the enterprise?

And then finally, mostly, from an educational perspective, but this really applies to everything. How can we use AI to enhance our workflows? So you can imagine there's a whole lot of sort of busywork that we all do in the course of our day-to-day work? How can we use AI tools to optimize that work so that we can do more meaningful stuff with our time? And that's my spiel.

All right. Apparently I'm going to have some computer issues over here...

It's the AI. Eric. It heard your talk. We were talking about it.

Well, we have about 10 minutes left, so we should be able to squeeze our remaining two speakers in here. Next up is James Fraser, to give us a bit of a research perspective. And he is also chair of the BTS department.

[James Fraser] Hi, I'm going to talk today just about how AI is transforming research perspectives. I thought it'd be useful to ground this in a case study of transformative success in my field of structural biology, which is the development of DeepMinds AlphaFold 2, which was really a game-changer in terms of its performance in predicting protein structure. So that's trying to predict the green cartoon on the right here with a blue model that matches it as closely as possible.

And so there have been many contests run for many years where performance is sort of baselined, flatlined, I guess, as you as you can see on the gray here, and then really AlphaFold2 gets us to this new performance level that's really a breakthrough, where it can be considered a replacement for structural biology in many applications, and certainly automates and gets rid of the boring parts of structural biology for a lot of other applications.

It's also spurring on transformative capabilities beyond its initial goals, especially in the field of protein design, where colleagues like Tanja Kortemme, and Bill DeGrado work. In here, I'm gonna show you a movie of a protein design process, using an AI method called "diffusion" that really takes very simple instructions, and allows us to create new proteins in silico, that often work in the lab at rates that are sort of unimaginable—were unimaginable just a few years ago.

So these successes really built on on a foundation of having: first, large and informative datasets, the first dataset being the Protein Data Bank, an open access archive of about 200,000 atomic-level structures of proteins. And so this provided sort of the ground truth to training AlphaFold; and the second is this clear, quantifiable benchmark, the idea of a winnable competition, that transformed this biological problem into an engineering challenge that attracted the attention of AI experts, leading to this model that outperformed all existing methods in protein structures.

So now that we've seen how this opened up new avenues of exploration, I think one of the interesting things that we have to ask ourselves is: How can we have an AlphaFold moment in other aspects of biology? And one of the key things is, is really thinking about the data-generation capabilities that we have at UCSF here—where could AI methods work, but they don't have the raw data to really be trained and tuned? And the other is in reformulating problems so that they have an engineerable element of it, so that the AI researchers can be attracted to making progress on these models.

I think that's one way of thinking about it, where it's a little bit transactional, between AI, researchers, and biological researchers. But I think what's most exciting is AI is becoming something that we embed really in every aspect of research. I think we're at a moment that's akin to the introduction of molecular biology in the late 70s and early 80s, into biological research—many of the advances catalyzed here at UCSF.

So if we think back to that moment, there was a lot of excitement about people who were discovering new restriction enzymes. But the moment that was really most exciting was putting restriction enzymes in the hands of people seeking to answer questions that had previously been unanswerable.

And I think we can have a similar-type explosion of progress, bringing that type of attitude to AI research, where even if you aren't trained as an AI specialist, it's now something that you have in your in your tool belt, much like you did you have molecular biology now in your tool belt. So I will stop sharing now to try to keep us on time and move on to the last lightning talk.

[Eric Davila] All right, thank you. Moving on to our final speaker, we have Jen Cocohoba, who is, in her vice dean role along with Kathy Yang, are tasked with working on clinical innovation and entrepreneurship. Tell us a little bit about AI and the clinical world.

[Jen Cocohoba] Thank you for this opportunity. AI is not new to pharmacy practice, given that much of what we do as pharmacy practitioners is grounded in the large amounts of medication-related data that exists in health systems. And I think we often take for granted the glimpses that we already see. For example, as a pharmacist, we are interacting with clinical decision support systems—the drug interaction popups, the little prompts that we receive to order pharmacogenomic testing, and when appropriate medication is ordered. Or even when you're typing in an order to get a medication ready to be delivered, whether it's on the outpatient side or the inpatient side, a little warning we receive saying, "Hey, that dose is not correct. It doesn't look right, please double-check it." All of those are kind of narrow, reactive forms of AI that we already interact with every day.

I think as AI continues to progress at UCSF, and just in general, the profession of pharmacy is definitely going to need to be ready to embrace any changes that allow pharmacists to practice at the top of their license. And this is that idea, that Dr. Sim brought up, of augmenting human cognition and really empowering pharmacists to do even more, more efficiently.

I think all of you probably know that [on] the School of Pharmacy faculty, there are a number of clinicians—clinical administrative folks—and we work very closely with our UCSF Medical Center pharmacy colleagues and other colleagues. And anyone who's going to be engaged in pharmacy practice will, at least on some level, have the opportunity to interact with evaluate, promote, or even challenge the use of artificial intelligence in our day to day.

So I think that would end up being the ask: to have our practice practitioners request access to the appropriate tools to be engaged in the process of trial and adoption of some of these systems and to provide lots of feedback on how these newer AI-driven systems improve or don't improve our patient care practices.

So if you're having difficulty trying to envision what that would look like in pharmacy, I by no means consider myself an expert in AI. I, apart from the pieces that I described, don't have extensive interaction with AI just yet, but it's coming and I anticipate that I'm going to have to at some point, be willing to look at it, learn about it just alongside everybody else.

So these came from our AI clinical care consultation group. Mackenzie Clark, Ashley Thompson, Allison Miller Pollack, Rebecca Kinnett, who is actually an external without salary faculty to us, Leigh Ann Witherspoon, Lisa Kroon, Desi Kotis, and Bani Tamraz.

But to give you some idea where AI can impact clinical pharmacy practice, really, we look at three large buckets, which is: 1) increasing the efficiency of how we dispense medications, and we have robots at our Mission Bay pharmacy to prepare or dispense, verify medicines—really managing all the inventory that we have with our medications—reducing waste, pharmaceutical waste, as well, and really managing all the challenges we're having with drug shortages, and being on top of that as well. There's ways we can use AI for HR and optimizing human staffing resources in the pharmacy.

But I think the biggest bucket is really allowing pharmacists to enhance their clinical decision making, and we talked about drug interaction checkers, but we also talk about predictive dosing, or optimizing medication based on specific patient characteristics, or predicting risk of adverse effects, so we can direct our pharmacist efforts into really caring for the populations who are most vulnerable, at higher risk—the highest risk—and meeting of the highest level of pharmacy care.

And there are a number of other patient-facing tasks that AI, I think, is going to impact in our future, or are starting to impact now. We're seeing chatbots for patient counseling, calculators, medication-adherence applications, and reminders and systems that try to improve the way that our patients use their medicines—synchronization, behavior change. All of these in some way will impact how we practice as pharmacists. And so I think there's some exciting things already in the pipeline, some things we're already using, and some things we may not even be able to imagine that are coming in the future.

[Eric Davila] Thank you so much for that, Jen. We're at time now. So there's only one outstanding question, and that's about resources, events, and such. We will provide that in the recap of this that goes in the faculty and staff news. You can find it in the faculty staff news section of the website, within a week or so. Back to you, Kathy.

[Kathy Giacomini] No, I just want to thank you, Eric, for hosting this interesting discussion. I want to thank Ida and all the panelists for their participation. And thank everyone for attending. Bye bye, bye.