Peter Thiel Explains Biotech Investing Rationale: Get Rid of Randomness

He can obviously evaluate the team, he can ask experts he trusts to give him their opinion about the value of the approach, but... biotech start ups, far more than software start ups, live or die by the strength of their technology. Transplants have far fewer false positives than working with cell lines (especially HeLa!) - but they are also several orders of magnitude more expensive. Is the trade off between decreasing false positives versus increase in experiment costs worth it? That company obviously think so, but it's not a no brainer, and it's very difficult to seriously evaluate. Mice models are very expensive - and they are better than HeLa, but they aren't foolproof. In cancer biology, there's a continuum between how expensive a technique is, and how much it recapitulates the actual biology of the disease. The sweet spot also varies hugely from disease to disease.

Another point which I find fascinating, is that more or less everyone who has worked on cancer biology knows that HeLa are not really a realistic model... yet people keep using them, why? Because a lot of researchers are interested in publishing papers in prestigious journals, not discovering robust findings. A pharma company instead has a lot more incentive in getting things right - because a splashy result that gets falsified at phase 1 is an incredibly expensive failure, while it's still a Nature paper for a professor. I don't think that academia should exist to serve pharma companies, but the problem of false positives is very real, and all the incentives around the publishing/grant system make it much worse. We should be rewarding researchers that get moderate but robust results that hold up under scrutiny, not people who make incredibly splashy discoveries that later turn out to be incredibly narrow and difficult to reproduce.

>They convinced me there is a surprising amount that has gone wrong with the cancer cell lines people have been studying

This sounds that they managed to explain to him how much is going wrong with HeLa cells (I assume he's talking about HeLa - these are the most widely used).

They're so sturdy that they contaminate everything to the point that some well-known, well-used cancer cell lines turn out to be nothing but HeLa contamination [1]. There's ethical problems as the original cells were taken from Henrietta Lacks without her consent (this is slowly being rectified, have a look at "The Immortal Life Of Henrietta Lacks"). HeLa cells have mutated so much in culture that they have little in common with human cells [2]. I wonder which "human cancer" cells they graft into mice.

All together, this sounds like the founders did their homework on science communication. To scientists this highlights the importance of communication - it's not "just" useful for talking to "the public about the importance of your work", it can make you some serious money!

[1] http://retractionwatch.com/2013/03/11/more-hela-problems-for...

[2] http://blogs.biomedcentral.com/on-biology/2013/09/17/debatin...

This is a major reason why consumer electronics and information technology have enabled so many more rapidly scalable businesses in recent decades than other industries (hence "We wanted flying cars, instead we got 140 characters").

The technology industry is uniquely founded on physical phenomena that are deterministic, even at high speed or recursive complexity (CPUs, storage, networking, displays, peripherals, etc). Therefore, if you can think of something in a mathematical fashion, you know that it can be built. This essentially eliminates technical feasibility risk - rather than worrying about whether a product can be built, you can focus on what should be built and why.

Derek Lowe posted about this a few days ago on his In The Pipeline blog:

But their rationale is that trying to culture CSCs in vitro runs the risk of having them change character too much, making any assays using them unreliable. My guess is that they’re right about that, but my worry is that xenograft tumors themselves are already unreliable enough to cause trouble (and I have no idea of what happens to them after you “passage” them through multiple animals). Xenograft models are, of course, well known in oncology drug discovery, but one of the things that’s well known about them is that they’re the pure example of “necessary but nowhere near sufficient”.

http://blogs.sciencemag.org/pipeline/archives/2015/09/09/cha...

This is a perfect example of why Thiel's thinking is so unappealing to me. First of all, nature is not random, but there is a lack of understanding of the true implications of cause and effect of what goes on at a biological (and ecological) level that it may appear that way to people so captivated by the binary control of the programming world.

But, even if it were random, I don't think that's evil in the least. It's unpredictable, yes, but so are the implications of many of the technologies that Thiel embraces. That does not make those technologies inherently evil, just as "randomness" is not an evil part of nature.

I am afraid of the sterile world that Thiel seems to want.

Additionally, I'd like to point out this bit from a below comment where the poster responded to several article excerpts, including: "I think of aging and maybe just mortality as random things that go wrong. The older you get, the more random things happen, the more breaks. If it’s not cancer, you could get hit by an asteroid. So on some level, technology is trying to overcome the randomness that is nature. That is a question on the level of a company. Can you get rid of randomness in building a company? But the philosophical version of the question is whether we can get rid of randomness in its entirety and overcome the randomness that I think of as the evil part of nature." Nature isn't random! Nature is complex, cryptic, self-referential, and chaotic, but it's deterministic at the level of biology. "Random things happening" is not a proper understanding of age-related decline; aging is the long-term compounding interest of metabolic byproducts. It really stuns me that Thiel's understanding of these things is so apathetic to reality.

Still, optimizing experiment design through a balancing of bayesian statistics, simulation and cost optimization may result in savings or less capital risk.

"The question is, can you change those probabilities into different numbers? The reason we invested in Stemcentrx at a valuation that would have been higher than many other biotechs we looked at is that we felt the whole company was designed to get these probabilities as close to one as possible at every step, to get rid of as much of this randomness or contingency as possible. That is something that we found deeply reassuring."

Broadly speaking there are two strategies for drug research: shotgun, and sniper rifle. If we're being honest, shotgun is the dominant method specifically because 10 chances is far safer than 1 "really good" shot that may or may not produce a really good drug. The one shot wonder shops usually fail early stage when they can't produce a chain of positive data from one funding round to the next. Biotech startups are longshots-- I would say that they are probably even longer shots than a software shop specifically because there can be unknown physical constraints of biology that can silently ruin an otherwise great idea, leaving other people to make the exact same mistake repeatedly after you're gone. Reducing inconsistency of your results is a core of the scientific method, practiced everywhere, but doing so effectively is hard when you don't have full understanding of any one component you work with. It isn't random behavior, it's just not fully understood behavior.

"One of the very unusual things they do is graft human cancer into the mice. It’s a somewhat more expensive way to do this than studying cancer in cell culture. It’s a somewhat harder structure to build. But drugs tested this way are much more likely to work in humans. They convinced me there is a surprising amount that has gone wrong with the cancer cell lines people have been studying. "

Literally everyone who is serious about making oncology drugs does this technique. It's a staple. I guess it's "unusual" if you aren't accustomed to the methods of biotechnology, but rest assured, this trick has been well-known for quite a while-- it is not a magic bullet, but it's better than alternatives. Cancer cell lines are known to be poor models in a number of dimensions, which is great because we can anticipate their shortcomings. Mouse models also have shortcomings that are somewhat well understood, but don't confuse this for a slam dunk. There are still many (most?) cases in which drugs developed from mice won't transfer to humans.

"But if biotech companies tend to invest money in ways that are pseudo random, then a lot of it must get wasted. You end up doing things where you say, “I am not sure it’s going to work.” Well, that sounds like a wasteful thing to do. The standard excuse that biotech companies have is that, “We don’t know if it’s going to work, so we have to do it this way.” That has to be inefficient."

Is this guy serious? You're never sure if it's going to work in the laboratory! Research is mostly "try something new, then try another new thing when that fails." It's "wasteful" because it's guided trial and error, and you don't have full totalitarian control over any of the variables. If you think you are investing in a new biotech company (or old, really) that has escaped this paradigm, I have a series of pyramids to sell you.

"This idea was very much in my mind by the time we invested in Stemcentrx in 2012. They had an annoyingly complicated problem, all these pieces you have to bring together, and they said, “We are just going to do it ourselves.” That is a mind-set that I very deeply share. I don’t want to name names, but there are other companies where, in some ways, this was the key thing that failed."

All biotech problems are annoyingly complicated, and the prerequisites for trying to solve problems are also annoyingly complicated. There are some extremely well funded companies that can do everything in house, but it's the wrong way to go about a startup-- even a lot of the larger biotechs contract stuff extensively. Why have problems with simple things that should work when you could contract them out and not worry? You can always bring that capability online in parallel in your own company at a later stage when you have more breathing room. It's a huge hassle to buy equipment, hire and train your own people to perform a task that someone else can do better and cheaper.

"The big picture is the question of whether biological science can be transformed into an information science. Can something that seems chaotic, fractal, and generally random be transformed into something more deterministic and more controlled?

"I think of aging and maybe just mortality as random things that go wrong. The older you get, the more random things happen, the more breaks. If it’s not cancer, you could get hit by an asteroid. So on some level, technology is trying to overcome the randomness that is nature. That is a question on the level of a company. Can you get rid of randomness in building a company? But the philosophical version of the question is whether we can get rid of randomness in its entirety and overcome the randomness that I think of as the evil part of nature."

Nature isn't random! Nature is complex, cryptic, self-referential, and chaotic, but it's deterministic at the level of biology. "Random things happening" is not a proper understanding of age-related decline; aging is the long-term compounding interest of metabolic byproducts. It really stuns me that Thiel's understanding of these things is so apathetic to reality.