So you want to start a Therapy Biotech
The experience of a software guy entering the fascinating world of biotech.
Over the past months, I have been exploring my interest in biotechnology, an in particular new therapeutic drugs. I had been looking at the world of biotech for a few years, but mostly from a software perspective, as my previous startup had Big Pharma as customers. I got the opportunity to dive into the fascinating world of actually developing treatments.
I come from enterprise software — I previously co-founded an IT infrastructure company, and then had a stint as a VC — so here I give the perspective of a pure software guy looking into the world of biotech. I hope this can give a refreshing, naive perspective and maybe some insights.
Everybody in VC will tell you that “biotech is a different world”: longer development cycles, riskier early stages, etc. Even more true for therapy. Great. Betting on early biotechs is more risky and more difficult. But that didn’t help me understand how to select the right bet.
So why is biotech so different? And will it always be?
Driven by Funding and the Legal Framework
Discovery, Preclinical and Clinical Development
The development of a new drug or therapeutic approach typically follows 3 stages. They all flow from the legal requirements needed to approve a new drug: every step is designed to gather evidence that will get the company to the next, and eventually enable it to file the regulatory requirements required to start out human trials and eventually approve the drug.
- Discovery and validation. This step starts off by a new approach being developed, sometimes in an academic lab, leading to an hypothesis. XXX approach, based on YYY (and ZZZ) new technology/findings, can help cure DDD disease. A company is formed. It will try to develop one or several compounds to exploit the candidate mechanism, select the one that is the most likely to work (the “lead”) using in vitro or animal models, then proceed to optimize it. Once it has reached enough evidence that this is a good candidate, it will enter the next stage.
- Preclinical. The lead(s) compound(s) is tested for pharmacodynamics (what effect does the drug have on the body), pharmacokinetics (how the body eliminates the drug) and toxicology (what adverse effects does the drug have). This is done first in vitro (eg. on cell lines), then in-vivo on animal models (eg. xenografted mice). The results indicate whether or not there is evidence that the drug can have the intended effect and is presumed safe for humans. It is used to file an Investigational New Drug (IND) application, that outlines what the sponsor proposes for testing in humans.
- Clinical: The drug is tested in humans, first for safety, then efficacy, and finally to show it offers patient benefit vs what already exists. Each step requires extensive investment, and paperwork. Succeeding in each step also tremendously increases the chances of approval of the drug, and the value of the company.
Early days funding
Therapy Biotechs are not expected to generate revenue long before inception, let alone profit. Their development is also completely constrained by the legal framework, as we have seen. Most software companies, post Series A/B could slash down cost and turn profitable, if their unit economics are sound. But they decide to burn cash to grow fast, conquer the market, and go big. Even the cash-burning, marketplace-like companies that require large scales and network effects to become profitable can usually prove profitability on a smaller unit of value (eg. Uber probably proved they could make a city profitable, before engaging in the massive cash burning growth phase).
Therapy biotech obviously cannot.
Therapy biotechs, at the early stage, are a scientific thesis backed by evidence and a team. The goal of the early stage will be to validate the scientific thesis.
A software startup is searching for a product market fit. A therapy startup is validating a scientific thesis.
That means that very early, aligning development on funding options is even more important than for other startups because there is no obvious other option. The scale of investment needed to cross the early discovery and validation stage or the preclinical stage also makes it inaccessible to angel funding. I have seen SaaS companies fail execution, go nearly bankrupt, and be saved by a small angel bridge: they cut costs, the leadership spent their time on sales and cleaned up the company then raised growth money when they were ready. This is very unlikely to happen with a therapy biotech, because an angel ticket will not be enough to reach a significant milestone, and there are no sales/growth costs to cut.
Early therapy biotech, in most cases, are also not suited to pivoting. They start off with a therapy mechanism hypothesis, backed by new science or experiments. This hypothesis takes a (few) million dollar to test out. The company raises money, builds a few candidates, and tests out the hypothesis. They may discover the assumption is wrong. Pivoting to a new mechanism is probably a desperate move. If the hypothesis is disproven, developing a new mechanism will cost hefty investments again and even if the goodwill of investors has not vanished, the founders have already started being diluted quite massively by seed and academic licenses for the science. The company’s best option is to cease operations: it has reached its goal (despite not with the hoped for outcome) in reaching a conclusion on its hypothesis. Atlas’ Venture Bruce Booth wrote a post about a real world (painful) story.
Except, if the (a) the mechanism was mature enough at the time of inception or (b) the company has a scientific platform.
2 Models / 3 Criteria?
This is a good way to systematically think about therapy biotech:
- Product-centric therapy startups are focused on exploring a novel therapy hypothesis, with one main lead candidate. Their development plan is focused on de-risking the hypothesis in the most capital efficient manner possible. Management is arbitrating capital allocation to experiments in order to achieve that in the shortest and most cost-efficient time possible. Protecting the IP around the main candidate is critical as it is the key asset of the company. Exemple: Arteus Therapeutics [story]
- Platform centric startups design, validate and exploit a scientific platform. Platforms companies develop a technology that can be used to develop a variety of candidates across several therapeutic areas. Their focus is both on developing their technological assets (“horizontally”), and advancing resulting candidates (“vertically”) to prove the platform value and value it. They oftentimes end up keeping a few candidates in-house (maximum value capture, high risk, high cost), and licensing out technology (low value capture, low risk, low cost) or candidates (medium value capture, medium risk, low cost). The textbook example is probably Genentech’s pioneering use of recombinant DNA technology, that led to the development of synthetic human insulin and countless blockbuster drugs (including a vast majority of Roche’s oncology portfolio). [story]
Discipline is key: building up a platform is a nice way to offer reassurance in case of the failure of a candidate. But good investors and entrepreneurs will have to determine if the platform is actually a disruptive and repeatable asset with many areas of applications, and not just an excuse to spend capital. Betting on a product-centric approach is scarier because failure can be on the path, with no good alternative. But it is also an efficient use of capital to pursue a clear, well-defined goal. In other words, and easier said than done: you have to know what asset you have, what you want to build, and not lie to yourself about it.
Judging startups is as much an art as a science, and every framework is debatable. When I was working in VC, the only thing I hated was people pushing for applying a standardized template framework to a company that was not fit to it. So I won’t make that mistake of professing a one-size fits all template. But mental models save time and are useful as a basis for first analysis. Plus understanding what is wrong with your model often leads to key insights aboutwhy this particular startup is different. So if I had to write down a compact way of analyzing biotech, I would use:
- Is the team experienced in the area they claim as their key technology?
- Do they have experience in drug development? Or do they have close advisors that do? Each step is costly, so teams don’t really have the option to figure it out along the way (in software you can).
- Do they have credibility in leading a therapy company? (If not are they clear on the fact that they will need to build or acquire it)
Platform (or IP for Product plays)
- In the case of a platform play (a) Is it repeatable, and to what extent? Is there potential for multiple therapeutic areas? (b) Platforms are hard to build and can get outdated fast, comparatively. (c) Is the underlying value broad enough to create lasting potential?
- In the case of a Product play, the main candidate is key. (a) What key finding led to the candidate? Is it solid and validated by multiple sources? (b) Is it novel enough (oftentimes novelty comes from mixing advances across multiple, previously distinct fields). (c) Is the IP protected?
Gauging the level of evidence required is tough, and from an investor perspective completely depends on the goodwill of the investor, usually driven by how proven the team is. For a company that builds a compound, is there a candidate already? What is needed to get there? Are the main points of failure clearly identified? Are the next steps identified, and are there the ones de-risking the company most efficiently?
Are new technologies changing the way we build biotech?
It is (somehow) changing
Software VC funding used to be very rigid: you had to prove a lot before raising funding, which limited the access to capital to experienced entrepreneurs with some cash to jumpstart things and VC connections. Putting something online required long development time and purchasing actual servers.
Then we saw a wave of change in IT that made the whole process easier. A lot of open source building blocks made it cheap to build things. The “as a service” model made it much simpler to get access to resources. Instead of purchasing a server, a big upfront outflow of cash, companies could get access to compute by the minute. Scalability was introduced. This transition is still taking place today, with computing workloads moving over to cloud, modular architectures, containerisation, etc. For startups, it means that getting started requires very little investment, and fundraising can happen once the prototype is working and product market fit is starting to be proven. If pre-seed / seed startups still raise early, it is probably more to get faster ahead than the competition than because they actually need to do so. All of that because units of value have become smaller: smaller amounts of compute have become available (renting by the hour instead of buying an entire server), software MVPs have become cheaper to develop, which has driven investment units of value required down.
As explained by Jared Friedman, this opportunity for small investments to jumpstart things is the opportunity YC identified and built upon. Jared argues that this model will come to biotech too. It makes sense.
Low-cost CROs are now offering affordable services, both in molecule development (eg. antibody development), lead selection (testing a few candidates for good properties before engaging in costly optimisation), and in-vivo animal trials. Developing an antibody can cost a few $10k. A few tests in xenografted mice can come as cheap as $5k / mouse.
All of those operations rely on CAPEX-heavy equipment: compound or antibody development requires complex screening equipment to be efficient. Testing on mice requires complex regulatory approval, trained staff, and facilities. So CROs are essentially doing to biotech what cloud providers did to software: instead of having to purchase a lab upfront (with facilities, a sequencer,high throughput screening equipment, trained full time staff) for millions of dollar, startups can now purchase experiments on-demand, as a service, at low unit price.
This has given rise to the “virtual biotech”, meaning biotech companies with no equipment, facilities, or staff except the founders. It’s worth noting that this change was more a business model change than a technology change.
Automation is also on the rise. Configuring and running servers used to be a manual work, that is now being automated through the DevOps paradigm. This is starting to happen in biotech, although efforts are still in their infancy. Companies like Transcriptic are automating the lab through robotisation and defining a standard protocol for experiments. This is so utterly fascinating that I will most likely write more about that.
But. Yes but. We are dealing with life science. The number of parameters are both huge and unknown. Learnings happen as much as the result of a planned experiment than as a side-effect of experiments (something unexpected happened, and it is actually a discovery).
Things take time. Even in a fully automated lab, mice take time to grow. Their supply and killing is limited by law and ethics.
Even if experiments are cheap, building a company will require multiple experiments, which will still cost money, and time. So yes, stuff is becoming cheaper, but not cheap. And it is most likely that early, preclinical drug development will still cost millions of dollars and not 1,000s of dollars.
And even if the very early stage becomes cheaper (work in progress, but slow transition), the preclinical and clinical stages, were the probability of failure is still (very) high) are completely bottlenecked by the fact that we are talking about injecting drugs into human beings. So the regulations are stringent, and doing no harm is the first criteria.
“Evolution, not revolution”..
I have been mentioning Atlas a lot in here, so let me point you to a fantastic article The Creation Of Biotech Startups: Evolution Not Revolution. With far more hindsight than I have, Atlas’ Bruce Booth delivers a few points.
Yes Biotech startups can leverage new models and CROs. But a few things are still true:
- cost: building biotech costs hefty investment (slowly decreasing)
- risk: the risk profile is still very risky until late in the company development
- regulation and ethics: regulation still drives the industry, and for good, as we are talking about human lives
- talent pool: experience is key (a SaaS startups can learn the rules of the game by trial and error, product and business-wise. Biotechs already have massive tech-risk, so they have to minimize other risks).
… Until the Google Cloud of experiments emerges
The way I see it, we are still early in this wave of transition in biotech:
- Costs are going down (slowly)
- the model of the industry has evolved from “everything centralized in big pharmas” to “VC money funds virtual startups that manage CROs then get bought by pharmas”
- automation is slowly ramping up
But the massive change has not yet happened. The big “cloud providers” of pharma are not there. Automation is juuuust beginning to happen. The technology stack is quite immature.
And when the technology stack is quite immature and there is no supplier out there, the best way to get access to the state of the art is to integrate in-house which means costs.
I think the model will evolve back to integrated biotech companies, that build capabilities in-house, bringing automation and AI in the picture, and outsourcing only a part of their experiments. Until automation is fully mature, the most-modern startups will rely on tightly coupled companies with a “drug development” arm and a “technology” arm, needing massive investments to deliver results, but both developing molecules and groundbreaking tech. Companies like Labgenius, that mix bleeding edge assets in many different fields, have the potential to change the game. Fully automated CROs like Transcriptic will take time to provide a service broad enough to significantly impact the industry.
But eventually, when the stack is mature, consolidated, automated, AI-driven CROs may emerge, finally unlocking “cheap biotech startups”.
This piece is coming from a software guy diving into biotech, so [unfortunately | I hope that] it will be obvious to biotech people. I am writing it for the benefit of people like me.
It was made possible by a great many readings and conversations, and was vastly inspired by blog posts from Atlas VC’s Bruce Booth and Bay Bridge Bio. Special thanks to David Sourdive and Franck Lescure for their invaluable time.
This is just the beginning of my forays into biology, stay put for more.
Thanks to Aldo Sanchez and Isui Aguilar for their time, thoughts, and patience. Thanks to those that helped me proofread this, including Alex Reeber, Maxime Voisin and other folks that will know this thank you is for them.
[Lebr] “Are biotechnology startups different?”
[Simu] “How to calculate the value of drugs and biotech companies”, Bay Bridge Bio.
[Vir] “The Virtues of Virtual — And Why We’re Devirtualizing”
[Plat] “Platforms versus products in the life sciences sector”
[PlatIP] “Protecting products versus platforms”