The Bio-Boom: Building a biotech company in Silicon Valley

As I drive up to San Francisco on US-101, I look to my right, and see large complexes with unfamiliar names – Mission Bio, Kezar Life Sciences, Ultragenyx. Drive into Stanford and you might pass Varian Medical Systems, Genencor International, or Stanford Genome Technology Center. The Bay Area landscape is still covered with large tech companies and countless tech startups, but the map is increasingly dotted with biotech startups. Big players, too, are constantly moving in. DuPont, in addition to owning Genencor, just opened their DuPont Silicon Valley Innovation Center in 2018 in Sunnyvale. Apple is making constant strides in healthcare – so much so that Tim Cook even said, “I do think there will be a day we look back and say Apple’s greatest contribution to mankind has been in healthcare.” And believe it or not, Google’s DeepMind is leading the field in protein engineering. 

A revolution is brewing, and the game is changing.

That much was evident at Stanford’s AI in Healthcare panel featuring a few of the faces at the frontlines of the biotech startup industry.

Dr. Tim Sweeney, M.D., Ph.D., is the co-founder and C.E.O. of Inflammatix. He began his molecular diagnostics company to target a surgeon’s worst nightmare – sepsis. Sepsis is a condition when the body attacks itself in response to an infection. With sepsis, speed matters, and he’s on a mission to create the fastest diagnostic test possible. The earlier sepsis is diagnosed, the greater the chance of survival.

Dr. Alice Zhang, M.D., Ph.D, is a Forbes 30 Under 30 Science Featured Honoree, and co-founder and C.E.O. of Verge Genomics. She is at the forefront of using machine learning to streamline the drug discovery process.

James Hardiman is a partner at Data Collective, an investment consortium for entrepreneurs in using big data to tackle big issues. He works with the board of several biotech and big data startups – Recursion Pharmaceuticals, Citrine Informatics, and Sequence Bio, just to name a few.

The moderator of this panel? He’s just the type of person you would expect – a businessman who has been there, done that, and still hasn’t stopped. Robert Chess, a serial entrepreneur in the life sciences, who co-founded Penederm, which IPO’ed and was acquired by Mylan Laboratories, is the current Chairman of Nektar Therapeutics, and is on the board of Pelvalon and Twist Bioscience

Chess began by asking the simple question, “Why did you start your respective company, what do you do, and what is its status?”

Sweeney’s interest began when he was training as a surgeon. He took time off to complete postdoctoral research in bioinformatics for post-operative infections. His study grew into Inflammatix, “a precision medicine company focused on new tools for diagnosis and response to autoimmunity,” he described. His company works by reading the expression of genes in the body, and thereafter applying machine learning to understand the variation of the expression of these genes. And, perhaps most boldly, he is targeting sepsis first, a disease many companies will not touch because it is so difficult to address. 

Zhang, similarly, saw a problem, and tacked it. Her company began because in graduate school, she was frustrated with the drug delivery design process, which she describes as “largely a guessing game, based on brute force screening.” At Verge Genomics, she uses machine learning to accelerate this process, with a particular focus on notoriously difficult-to-treat diseases like ALS, Parkinson’s, and Alzheimer’s. She has 20 employees now and expects that to be 40 a year from now. 

Hardiman, an investor in these types of companies, explained that many companies forget that proprietary data sets are the most compelling part of a company built on applications of artificial intelligence. When a company claims to have a science or engineering breakthrough built off of a public data set, Hardiman approaches the company with some degree of skepticism. Zhang, for example, produces all of her own data in-house, and thus has the exact type of proprietary data that Hardiman finds attractive in a company.

Data is crucial. The critical step that enabled researchers to generalize classification of skin cancer was the use of 129,450 clinical images, whereas previous research was based on as few as a few thousand images. The point Hardiman expresses is a valid one.

Chess followed up with, “Why develop an actual product when you can just stick to machine learning and artificial intelligence platforms? Why have these mixed models?”

Zhang’s route to product was indirect. “When we started, we said, ‘Let’s build a platform for drug discovery!’ and began to think in terms of service organization. But then we started to get exciting results that showed that our approach was working. An experimental drug even reversed paralysis in mice with ALS. That’s when we realized that our value is not in selling the service, but in a potential billion-dollar drug. If you have conviction in what is coming out of your platform, then pursue that. In fact, our hardest decision early on was walking away from guaranteed capital in the bank from term sheets with pharmaceutical companies. We instead decided to develop a product, which involves significantly more risk.”

“We had a very similar experience,” Sweeney explained. “We tried to develop content and a diagnostic platform, and we even partnered up with a startup, but unfortunately that compounded risk of a new product and an unproven platform. Our early stage research was more successful than we had imagined, and we decided to pursue the product route. This involves a lot more time, a lot more risk, but is much more profitable.”

“Agreed,” Hardiman added. “The way to capture value in the space is to capture an asset, and most investors, if they believe in the asset or platform, then they will want to fund the company. Traditional biotech and healthcare investors are more interested in a single asset risk. These are largely M.D.’s and Ph.D.’s who believe in the asset. Whereas tech investors, who are starting to glean over into the bio-space, are more interested in platforms and the tech side of things. There can be tension between the tech investors and the traditional biotech investors. Importantly, if a traditional biotech investor wants to come in, then that is a strong signal to the market that the biology and science coming out of the company is real.”

“Great, this leads into my next question.” Chess asked, “How have you thought about investors and how much do you want tech investors versus single-asset traditional investors?”

“For our business, this was a 50/50 split.” Incredulously, and with a slight chuckle, Zhang noted, “I had two different decks for the whole fundraising process. For the tech investors, I added terms like ‘network effects,’ and the users were ‘drugs’ – they get it. But the life science slides were cut-and-dry. This is the target, this is the biology, this is 1.5 years away from clinic, we’re in phase 2, etcetera, etcetera. That was how we successfully managed the process. There is just not yet a group of investors that represent these types of companies. In our own process, it was definitely apparent that got both investors interested. The cultural differences are very different. Tech investors are generally faster, more confident; life sciences are very collaborative – they might even ask who else is leading your round!”

A round here refers to a series of investments, usually by venture capitalists. Asking who else is leading your round might be intrusive depending on the circumstances.

Chess turned to him and asked, “Does this layer on another level of complexity as a diagnostics company? In diagnostics startups, there have not been a ton of successes. Bread-and-butter molecular diagnostics companies have been very difficult to fund over the last decade. How did you fund your company?”

“As a machine learning plus health platform, we wanted a mix of both healthcare-focused and tech investors. Vinod Kholsa led the first round.” Vinod Kholsa is a big name in Silicon Valley. He is the founder of Kholsa Ventures, who has funded many tech companies such as academia.edu, DoorDash, and HackerRank, among many others. “Building a company in diagnostics has been difficult in that you have to show value quickly. You have to build something, get regulatory approval, and then get reimbursement before you really take off. The short answer is that we started with someone who could appreciate that we had a novel approach and could overcome the technical barriers to success.”

Chess then posed the question that many have considered regarding fads: “If you look at ML and AI applications of diagnostics and therapeutics, it is getting a ton of money right now. But so have other things, such as functional genomics and rational drug design. Millennial was the leader in functional genomics, but they never got a single product even into the clinic. Combinatorial chemistry? Same thing. None of these things turned out to be as big and as quick as anyone thought they would at the onset of things. Why is this different?”

Zhang quickly presented a thought-provoking insight, “I think there has been, with any tech, whether it is gene therapy or genomics, two waves. The first wave figures out all of the complexities, all of the nuances. The second revolution, about 10 years after the first, is around the business model. The question is asked, ‘Having seen the struggle, how do we adapt?’ From the AI and ML perspective, our view is that the power and ability to generate massive amounts of data to accurately reflect disease biology has also changed. We can do that internally and actually test those predictions and feed that back in.”

Her thoughts are reminiscent of Reid Hoffman’s book Blitzscaling, which considers how to rapidly build massively valuable companies. Reid explains that the entrepreneur needs technological innovation, strategy innovation, and business model innovation. Here, the speakers explain that it appears that before the second wave can take off, the technology must become further refined, business models adapted and innovated, and the strategy for going to market – product versus platform, physician-focused versus tech-focused – revised as well. Hardiman and Sweeney agree that the time is ripe.

“This wave is different,” Hardiman stated. “For most of history, biology has been reductionist. The cell, for all of its complexities, is largely beyond human comprehension. The hope is that AI can capture a lot of that complexity, and hopefully we can predict biology better and then this will be different.”

Sweeney added on, “An ML and data approach is necessary, but not sufficient. All of the new products are fundamentally about machine learning. We have seen a lot of failures because there was not a focus on building a product focused on making the patient’s health better, or that a physician wants to use. Would it be something that every physician used every day?”

Chess focused his attention on Hardiman: “Now, your firm has based your thesis around this. What areas will monetize the quickest?”

Hardiman responded, “I still think therapeutics, for the most part. People tend to pay for treatment before they notice something is wrong with them.”

Chess then asked the group, “What are the public data sources available and what are the underlying biases and problems with these datasets?”

“In nucleic acids research, there is a list for biology databases available publicly – hundreds of them. If you are not a domain expert, you can quickly get led astray,” Sweeney warned.

“Well, if you publish using NIH funding, then you have to upload to a database. There is a ton of free data for chemistry. For our purposes, we made use of pretty basic discoveries, such as GWAS [genome-wide association study] studies and mutations that predispose those to disease,” Zhang explained. “But here’s the thing: We first started to work with public data, but we quickly moved away. Often the people publishing the study have very different goals than you do. For example, they might just publish something to validate a gene, and so their sample size is ten, total. Ten! It is very noisy, there are many variables and it is hard to extract data. That’s why we produce our data in-house.”

Ten is way too little, and it is very difficult, if not impossible, to produce statistically significant results based on a dataset of size ten. If there is not a centralized database for the type of information being analyzed, then researchers who rely on public data might not have sufficient data to produce meaningful results.

Hardiman gave his personal perspective. “As an investor, we are generally skeptical if you say you made a discovery with a public dataset. Unless you have property, most of the value is having access to proprietary data. If you do have access, then it becomes very clear why you have discovered something someone else has not. If you use public data, then you have to be able to answer questions on why public data is meaningful and interesting.” 

“Now, at the end of the day, you’re building a company.” Chess transitioned. “Everyone has a China strategy these days. James [Hardiman], you encourage companies to look internationally. It might be too early, but have you even considered that?”

A China strategy is a path for a company to expand operations in China. This is critical because China both has over a billion people and a burgeoning middle class.

“Our second biggest investor is WuXi AppTec. They are a prolific investor in China as well as have venture arms in China. Lots of biotech companies look for Chinese investors for bigger routes,” Zhang said before she passed the microphone to Hardiman. 

Hardiman followed this up by saying, “Most companies I invest in are very early. Starting in the United States and China is a recipe for disaster because you want to focus on tackling and winning one market. Generally, this is better for Series B or later. On the therapeutics side, you have to ask: ‘Are you developing a drug for a disease more prevalent in certain populations?’”

“OK, last question before our lightning round. Did any of you two go to an incubator or accelerator?”

Incubators and accelerators are both programs for growing companies. They often receive some portion of equity for the program and monetary investment, but not always. Many of these famous programs now have a wing dedicated to biology and the life sciences, such as Y Combinator’s YC-Bio. 

Zhang began. “YC [Y Combinator], and it was a profoundly valuable experience because of the network it provided me. Such an incredibly powerful network of other founders. It was how I found out what executive coaching was. It is incredibly powerful as a company getting off of the ground. I have not heard of another incubator that offers as much as YC does. Even some of my best friends are from YC!”

Sweeney had a different approach. “I did not. We joined StartX, but we did so after the first found of investment closed.”

StartX is a famous organization that is neither an incubator nor an accelerator, but rather a nonprofit dedicated to helping founders thrive.

“OK, lighting round. Answer these questions briefly. What is the most overhyped area that is getting more attention than it should? What is underhyped, under the radar?”

Hardiman sighed, “AI for replacing the radiologist is overhyped. Dozens of companies are working on this and I think it is difficult to make a product compelling for end users. You can build it, have some data set, some image modality, and then the models are specific – CT for lungs, MRI for heart.  It sounds nice which is why many have tried to tackle it. AI certainly has an ability to make the radiologist but efficient, but I don’t think that that’s the right place. Maybe in image capture itself, such as making an MRI image take 20 minutes instead of 40. But definitely underhyped is computer vision and biology. A lot of biology is slides and cells.”

Zhang was also passionate. “If one more person asks me about the gut and the brain, I am going to scream. It is pretty overhyped because the studies are not super well done. But! There is a ton of work to be done in bettering clinical trials using AI wearables. There is some hype surrounding this, but the area is within reach to drastically improve it.” 

The quote by Tim Cook mentioned earlier comes back to mind.

“Early stage cancer detection is extremely overhyped,” Sweeney, CEO of a diagnostics company, speaks from a wealth of experience. “These companies have astronomical valuations and so far, there is no proof. I have been talking to some folks that do this, and there are lots of technical problems that have not been overcome, even theoretically. However, the thing that jumps out to me as underhyped is the ICU and acute care. The ICU is by far and away the most expensive part of the hospital, and one that is ripe for innovation.”

Chess put them on the spot again. “What is one area that will be radically different ten years from now?”

Sweeney, who is currently targeting sepsis, naturally considered the operating room. “The operating space will be very different. As a surgeon watching some of the innovations, watching some of the autonomous robotics and procedures show really, really good efficacy. I am not sure about the ten-year timeframe, but it is certainly coming.”

Zhang, who is developing drugs using her proprietary data and technology, explained that clinical trials will be transformed. “It is just mind-blowing how they are done now. The whole study can be thrown off by the slightest thing. I do not imagine them being the same in ten years.”

Hardiman wrapped up the question with his usual, more modest approach. “It is hard to know. I believe that it will largely be the same in ten years. But if you wanted to build a company rapidly, you can do a lot if you touched how money flows and how risk gets transferred. For example, the way a lot of physicians get paid. There are these codes and a human who reads the notes and code submissions to insurance company – you do not need people doing that. Relatively constrained text, a set of constrained CPT codes – totally doable. Things that touch how money flows and how risk transfers offer the opportunities to really drive change.”

But that is what unified them, all four of them, including Chess: the chance to drive change. To make their mark. Better the world. They came to the right place, took a risk, and are fighting the good fight. 

By giving you some honest advice, they hope to offer a tiny light for your journey to drive change. But for now, they’re off to the races, and they’ll see you there.