Faculty Interview: Brandon Allgood

Brandon Allgood is a serial entrepreneur and researcher focused on applying machine learning (ML) and large scale computational methods to improve human health. Currently, Brandon runs a consulting and early stage investment company, Obsidian Scientific. LLC, where he uses his 19 years of experience in developing AI based platforms in healthcare to help large pharma, venture capital, private equity, and start-ups with AI strategy, platform architecture, early research, and technology management. Brandon previously helped build three companies, Valo Health, Numerate, and Pharmix, at the forefront of applying modern ML and computational methods to drug discovery and development in diverse subfields, including chemistry, biology, clinical trials and real-world health data. Brandon received a B.S. in Physics from the University of Washington, Seattle, and a Ph.D. in Theoretical Cosmology from the University of California, Santa Cruz. Brandon has authored scientific publications in astrophysics, solid-state physics, computational biology and chemistry, and business. He is also co-founder and former executive chair of the Alliance for AI in Healthcare (AAIH), former UCSC Foundation Trustee and current UCSC Baskin School of Engineering Dean’s Council member.

You came from a background in physics and theoretical cosmology and transitioned into
computational biology, data, and healthcare, could you walk us through how you ended
up on this career path?

When I was young, I was interested in mathematics, but did see a career in that alone. Physics was introduced to me in a very quantitative way compared with the way chemistry or biology were taught, and I was immediately kind of drawn to physics because it aligned with my interests. Within physics, I gravitated to the most interesting things from a mathematics perspective—cosmology, general relativity, and quantum field theory. In graduate school I spent a lot of my time writing software to run simulations on big supercomputers. Ultimately, I wanted to be an academic, but towards the end of graduate school, I became disillusioned with the idea. Being in the Bay Area I turned toward Silicon Valley for industry work. I looked at Google, I looked at NASA, and I looked at a small startup company doing some interesting things around machine learning in chemistry. What Google did was serve ads, and that didn’t interest me. My interest is in science and math, and NASA felt a little too isolated. So, I decided to go to the startup that was doing chemistry because I figured chemistry is just physics, and I know physics and math, so I’ll go into that.

I learned very quickly that I didn’t know how to code as well as I thought, but I shared an office with an amazing software developer. I would say he destroyed me daily, which turned out to be good and helped me improve my ability. Being in a small startup, I also got to see what it was like to build a company, to make mistakes—you know we made a lot of mistakes—and how to do lots of different things. It was something that I enjoyed, that we could do cutting-edge science in a small company. It was just exciting. From there on out, I went from chemistry, expanded to computational biology, and then I started working with human data more recently, such as electronic medical records. It was more of an evolution over time where I followed what was interesting to me at the time.

Do you have any advice for students navigating interdisciplinary career paths?

I would say if you’re jumping careers or jumping across boundaries, you’re going to have to come up with something that makes you stand out or something that gets you noticed. Some of those things might be trying to volunteer in that field or work on an open-source project because won’t generally people turn down free work. This way you get expeirence and they get stuff done. So maybe there’s a place to volunteer to get a little bit of experience. Academia is a good place for that, for example, volunteering in a lab. If there’s something that you can do at your university or something you can do for free, that can build up that experience.

Another piece of advice for resume building is to focus on the work you have done, not your degree. I work with a lot of students and young people just getting started in their careers. The thing that becomes disappointing to them is that while they are super proud of their degree, it should be put at the end and not at the top of the resume. It feels like, “I just spent how much money and how much of my life getting this degree, and they want you to put it at the end of my resume?” Yes, unfortunately, because in many cases that’s just your ticket to get in. What we care about as employers is not your degree. I mean, yes, if you’re going to come and work as a biologist, I expect you to have a biology degree, but what’s more interesting is what have you done? Talk about your final projects in your classes. Maybe you’ve got a capstone or an internship. Write about what you did. Be very specific about what problem you solved and tell us what impact that had. Don’t say, “I took this course, this course, and this course.” You can list projects, and in the project list give what course each was done in. If you’re applying for a position, make sure your resume fits the position. Also, don’t just have one resume that you send out to every position. Actually research the company you’re trying to work for, understand what it is they do, and then maybe change the verbiage and move some things around in your resume. Put the things at the top of your resume that you think that that company is looking for. It’s extra work, but I think you have to do that. In your letter of intent, don’t talk about what you’re going to get from the position. As an employer, I don’t care what you’re gonna get. What I care about is what you are going to bring to my organization. As a potential future employee, talk about what you can do for the employer. Yes, you’re going to learn some things, and everybody’s going to learn some things, and that’s great. But it should be about what you’re going to do for them.

What's your advice for other entrepreneurs?

Berkeley, UCSF, and Stanford, all have really good networks and connections for entrepreneurs, and you should take advantage of that. I’m involved in the networks at UC Santa Cruz, which was my PhD alma mater, and the University of Washington, which is my undergrad alma mater. I think entrepreneurs should get involved in entrepreneurship networks around their university or think about joining incubators. One person is going to have an idea, and maybe it’s a business person that needs a technical person, or maybe it’s a technical person who has an idea, and they need the business person. No company was ever started by one person. Go out and try to find the person that complements you.

One of the things that I find somewhat frustrating is most people don’t ask for help. Somehow people feel like they should do it on their own and they should never ask people for help. You look on LinkedIn and you see somebody that you admire, maybe a vice president somewhere or a CEO somewhere and you go ‘Well, if I try and send them a letter, they’re never going to respond.’ I would ask you, what’s the worst that could happen? If you said, look, I’ve always been very interested in technology, so I’m gonna write to Satya Nadella, the CEO of Microsoft what’s the worst thing that could happen? He could ignore you. Okay, you got ignored, no big deal. But one thing you’re going to find that’s surprising to many people is that most people that have been successful in their careers, want to talk about what they’ve done, want to share their experiences, and want to help others. I don’t have a lot of time, but I love helping people, and I think most people do enjoy helping others, especially younger people starting their careers. So, reach out to people you want to learn from.

What inspired you to start teaching again?

As I said, I have wanted to be an academic since I was 12. I left academia due to some of the financial aspects of becoming an academic, to be completely honest. Doing the post-doc rounds meant dragging my family around the world which was challenging. My plan then became to go out into industry, and make my fortune, if you will. Then once I didn’t have to worry about those kinds of things, come back to academia. I’ve always remained involved in academia since leaving. I was on the foundation board at UC Santa Cruz, and I’m still on the Engineering Dean’s Council at UC Santa Cruz. I’m currently teaching at Berkeley. I am involved in entrepreneurship at UW. My plan was to eventually find my way back in because I’ve always enjoyed teaching. I’ve also always enjoyed doing pure research. So, teaching at Berkeley is part of my foray back into academia.

During your career, how have you seen computation changing the pharma and
healthcare industry?

In 2005 I would walk into a pharmaceutical company and say we’re using AI, and they would say, “What?” Now you have every single CEO of every single pharmaceutical and biotech company asking what we’re doing with generative AI. Everybody wants to know, so it’s flipped completely from the question, “What is AI?” to the question, “What is every single individual in our company doing with generative AI?” In the beginning it was hard to get people to understand the power of what machine learning and AI could do for drug development, and how it could help patients. Then there was this kind of flip about 10 years ago, where I think we built up enough proof points around how machine learning and AI can positively affect drug discovery and drug development. There’s been this transition to where now the questions are how to use these tools, when to not use them, and how they work within the organization.

In pharmaceuticals, we also must think about how best to use human data. We have to be very careful with people’s privacy and make sure that they have consented to have their data used to train a model. We also need to be thinking about things like equity, equality, and bias, and we need to make sure that the datasets that we’re using represent the actual patients with those diseases, and that we’re not just using data from rich white patients to train models. That also means that there needs to be more transerancy and better communication around the use and benefit of letting your data be used.

I also think that the pendulum has swung too far in the wrong direction. It’s overcorrected. Now everybody and their grandma wants to know what we’re doing with generative AI. But at the same time, they don’t fully understand it enough to understand what their question means. So, while I love the enthusiasm, I think often we’ve got to dial it back and concentrate on some of the fundamentals. I would say a lot of things have changed, but I think we’ve still got a long road to go for AI and ML to have the impact that I think it can have in drug discovery and development and human health.

Where is AI today the most successful? AI today is most successful in places where we as humans already know the answer. For Example, if an AI says, “Look, there is a zebra in this photo,” a five-year-old can look at the same photo and say, “That’s not a zebra.” We already know the answer to the question, and what AI allows us to do is to generate new versions of that answer. But the thing is, we know more about how the galaxy works than how our own body works. With biology, AI is extrapolating in the dark. With a generative model that’s trained in biology, it is trying to predict a connection between a gene and a disease, or predicting whether a small molecule is going to inhibit an enzyme. We don’t already know the answer to these questions, so it’s a fundamentally different space from ChatGPT. No one knows the answers to these questions, so we’re always kind of extrapolating away from the original training set which makes it a very unique problem and also makes it more challenging.

What developments do you see in the future of AI and machine learning in the health

We as an industry have to organize our knowledge better and collect better datasets. It all comes back, in most cases, to having better training sets to train the models. Biology and to some extent chemistry are unique sciences, because we’re studying things that are so complex, and not amenable to reductionist thinking. Physics is a reductionist science, so you can reduce it down to the base concept, and then from there, build everything back up. In biology, we sequenced the human genome, and I remember when we thought diseases were done since we’d sequenced the human genome. It turns out, no, it’s way more complex than that. There are all these feedback loops and everything. With biology, I think we as humans still have to organize our knowledge better, develop better training sets, and then train models that can begin to make better predictions. So, I think there’s a lot of work to be done, getting good clean datasets, getting datasets that are representative of humans and not animals or cells in a dish, and not just from narrow demographics. I think most of the work is not on the AI side—it’s more on us as humans getting the data into a good place so that machine learning can take advantage of it.

Overall, there are going to be some jobs that are taken over completely by AIs. But in most jobs, it’s not that the AI will take the person’s job, it’s just that people who know how to use AI are going to take the jobs from the people who don’t. It’s a very powerful tool, and it is incumbent on every person to learn how to use it best. It’s going to be a skill that will be required in job descriptions. An example of this is prompt engineering which is the skillset of being able to properly handle ChatGPT prompts to get the correct answers out. It’s not trivial. You have to learn its idiosyncrasies.

Could you tell us what students learn in your course, machine learning algorithms?

Students learn the mathematics behind optimization and machine learning and how to actually write the code required to build and use models. In the lectures, we work through various aspects of machine learning starting with Optimization. We then talk about how optimization leads to the minimization of functions, which can then lead to machine learning. We talk about supervised machine learning and unsupervised machine learning, and later in the semester we talk about generative modeling and recurrent learning. We try to give the students a breadth of various aspects of machine learning that gives them a good base to begin their careers from. Students will also learn the software skills to be able to code up algorithms to understand how those machine learning algorithms work. Because it’s in the MSSE program, the examples that we use are chemistry, biology, and protein-based.