Or:
Biology is really hard, and returns to investors in the space have been pretty tame, and taken longer than promised (expected). See commentary by Peter Thiel in Zero to One for more on this topic.
In response to this reality, one tactic is to create a narrative that sells the vision of biotech as a tech or engineering discipline to an investor audience that is naive about biology and uber bullish on tech eg, this article).
"Biology? It's just engineering, we can engineer things, we'll just engineer biology (please invest in us as if we're tech]".
Never mind that engineering is a tradition built on minimizing variation, while biology is a science founded on understanding a built-in mechanism for generating variation.
"Life finds a way". Indeed. I've been on numerous synbio projects and one thing is certain, evolution is always driving against the engineered design. And by driving against, I mean undoing. Entropy might slowly undue engineering, but mutation and selection? It's an entropic accelerant for engineered outcomes.
Also, engineering is about "parts" and synbio is constantly dreaming of building things by putting parts (promotors, genes, etc) together from a catalogue. In reality, biology is interactions, where the expression of almost everything is dependent on the context of all the other parts. So outcomes aren't simply "the sum of the parts". When they are it's for a minority of cases.
Finally, as you scale a process (say from a test tube to a 1,000,000L industrial sized fermentation tank, biology almost never cooperates.
Biotech's dirty secret is that biology is Not engineering (it is still a discovery science)...but the music is still playing folks, dance on!
(Sorry, it's early, I'm still pre caffeinated, and this topic always rankles me).
This seems a fairly narrow view. In my field (oncology), there are many drugs saving lives that are the result of what is engineering in biological systems. Certain immunomodulating antibodies are the most recent example, one of which is earning it's pharma company many billions of dollars a year. Another proven class of drug is conjugating chemotherapy to the back end of an antibody. These drugs have undergone incremental improvement, with the latest versions showing greatly enhanced efficacy. This to me meets the criteria for engineering. Then there are CAR T cells. Researchers in this field actively describe their work as 'immuno-engineering'.
Synbio has been disappointing admittedly, but then engineering an organism is orders of magnitude more difficult than any engineering problem in existence. It is bottlenecked by the lack of a truly useful DNA synthesis platform. One day, when you can dial up a genome and have it into a cell in < 24 hours, and do it in a high throughput fashion, things will be very different. This will happen in under 10 years I think.
The top selling check-point inhibitors ie "immunotherapy", Keytruda and Opdivo, were not found by anything remotely resembling 'engineering'. It was a traditional lab science and drug discovery. If anything those two drugs have expanded their lead commercially, with no other PD1 drug approaching it. None of the top 10 selling Abs are drug-conjugated. CAR-T is struggling mightily in the clinic - it is the opposite of "scalable" due to the personalized nature, and has a terrible side-effect profile.
TCRs are far from 'engineered' in any real sense, highly dependent on patient MHC and the specific neo-antigen, with no one able to rationally design from scratch a TCR.
Synbio's challenge is not necessarily engineering a genome faster. It's that no one knows what to make, per se. What should we print? We don't know how to design proteins or enzymes. Designing assays to measure a good vs bad design is often the rate-limiting step.
I see you have a specific idea about what ‘true bioengineering’ should look like. But you have just made this up, based on your own ideas, which I simply reject.
Your first paragraph glosses over all the work that goes into making a useful antibody candidate, a process of iterative modification to meet certain specifications. I don’t see a problem calling this engineering. Your point about market penetration as a judge of ‘engineeringness’ is also silly. There are antibody drug conjugates that serve a specific patient population very well, the fact they aren’t in your ‘top 10’ is irrelevant.
So what if CAR-T cells are struggling in the clinic? It is an entirely new treatment modality. There are people walking around who would be dead without it. You deride it because it isn’t an instant global panacea?
Your point about Synbio is also wrong. Leaders in the field just wrote a position piece in Science and genome writing was listed as one of the major challenges. You’re trying to tell me that George Church ‘doesn’t know what to build?’ Pure bunk.
Dear Gatsky, My point is not that there is no engineering or no biotech in the world - to the contrary. There is antibody engineering, biomedical engineering, and many other types of engineering. I am a computational biologist. I actually worked at the George Church Lab, playing with the Polonator (an early sequencer prototype). I started a computational biology company (Seven Bridges Genomics), and then spun a biotech from it. So I truly believe in engineering.
Try to think of my argument as trying to add higher resolution to the debate, instead of being against/for your argument.
My main argument is against over-crediting what the software and algorithmic component of this is, precisely because I think it is too important. Another wider argument is are humans rationally designing anything / do we understand the wider design principles.
So let's go back to immunotherapy. Neither the target PD1/PDL1, nor the therapies themselves were enabled in particular by "in-silico" methods. Nivolumab and Pembrolizumab were not designed with much algorithmic or software input, but discovered through good old experiments / lab science/engineering. There is no human designer or algorithm in the world where if I give a target, they can rationally design an antibody. What we do is brute force + affinity selection / "directed evolution" to force nature come up with an answer still. We have no clue how to design it like one would design a bridge.
I think CAR-T is amazing technology - my point there remains: the therapies so far did not at use in-silico design of the Chimeric Antigen Receptor so far (the CAR part).. The low success in the clinic is precisely due to our inability to actually design these things properly. Similarly, neoantigen selection remains a big challenge due to lack of rational principles. A CAR is not too different from a TCR or Antibody, which we don't know how to rationally design.
I believe that genome writing is a major challenge. Again you are widely missing my point: Suppose you had a very cheap, abundant, perfect genome writer. What would we write? George, who was an advisor to my first company and kicked off our Hackathon, would be the first to admit that protein design is a major challenge.
Venter wrote an entire organism - and it was mycoplasma verbatim.
We are only able to copy nature or cajole it at the moment, because fundamentally we don't know how to design biology. It is because we don't really understand it very well, which makes it such an exciting time to be in the field.
None of this means these won't be done, or that teams around the world are not trying to do it this moment. There are amazing labs and companies working on these. What I am saying is more specific: Immunotherapy, CAR-T, or genome writing, in its first generation, as it exists, cannot be recruited as a showcase for computationally led, "in-silico" driven advances in biology.
I agree that great advances in biotech have come from bioengineering (all the way back to Genentech, the first biotech company) and I think that a lot of great advances in biotech will continue to come from bioengineering.
But the narrative that "biotech is becoming more like software" basically ignores the fact that engineering's impact in biotech is limited by our understanding of biology. The biggest challenge in drug discovery is that most targets arent good, and most disease models and in vitro systems are poorly correlated with clinical outcomes
You can engineer all you want, but if you are engineering a drug to optimally bind a target that isnt actually important for disease, all that engineering is wasted. Big pharma companies thought they could "solve" drug discovery with engineering / efficiency improvements in the early days of high throughput screening, but that only works if your biology is correct
The exciting thing about checkpoint inhibitors for cancer is that their targets are actually really important in cancer's strategy to avoid the immune system. The engineering work required is very impressive as well and built upon decades of innovation in that field, but the big pharma companies had a lot of that engineering expertise already I believe. Once they learned the importance of PD-1 / PD-L1 they could execute development campaigns using their existing infrastructure
>And that reshaping is a pretty lively process. Witness the news that Roche has had a major unexpected clinical failure with their antibody therapy Tencentriq (atezolizumab) in bladder cancer. This is another antibody targeting the PD-1/PD-L1 system, like Keytrude, Opdivo, etc. (in this case, it’s going after PD-L1 itself), and it was the first one approved against any form of bladder cancer. It’s also been approved for metastatic non-small cell lung cancer, which is a more traditional indication in this area, if by “tradition” we mean the last couple of years. The bladder cancer approval, though, was an accelerated one after just Phase II data, with re-evaluation to come after the Phase III numbers came in.
>Well, now they’re in, and the drug missed its primary endpoint. This not only puts Tencentriq’s continued approval for this indication in doubt, but it cannot be good news for the other companies in this space who are targeting it as well. Opdivo (nivolumab) from Bristol-Myers Squibb got approval in February, Pfizer and Merck KGaA’s Bavencio (avelumab) got a similar accelerated approval just a couple of days ago. Merck’s Keytruda (pembrolizumab), meanwhile, showed a good response Phase III (so much so that the trial was cut short), and is under review at the FDA, and there’s AstraZeneca’s entry in this area Imfinzi (durvalumab), too.
>But a similarly targeted drug from Clovis Biotechnology, rociletinib (which is one of the recent acrylamide-containing irreversibly covalent kinase inhibitors), ran into some big trouble. The FDA wants more data, for one thing, and the reason that they want more data is that Clovis submitted preliminary clinical data to them earlier that have not held up. That was clearly done because they were in a race with AstraZeneca (and with time in general), but you’re walking on – or sprinting across – a flaming tightrope when you try something like that, and the results are clear. Clovis’ stock fell about 75% on the news
>A closer look at the data, though, tells an even more different story. That overall POS figure is heavily dragged down by low success rates in oncology. Of the 41040 total pathways in the set, 17368 are for oncology (note that the same drug tried against two different types of cancer will show as two different pathways). The POS of everything outside of oncology is 20.9%, which the POS in oncology itself is 3.4%. If you look at lead indications, instead of all indications, the POS goes up overall (which is in line with earlier studies). But the Phase 2 to Phase 3 transition rate actually goes down a bit, interestingly. Oncology is still the lowest of bunch.
>The timing of this report from the FDA is surely no accident, but it’s always a good time to think about this: the great majority of all drugs that enter clinical trials fail. They fail because they don’t do anyone any good, or because what good they might do is outweighed by some serious and unexpected harm. Around 90% of all compounds that start in the clinic never make it out. Even by the time you get to Phase III – and these are drugs that have apparently already worked in sick patients by that point – the failure rate is still nearly 40%. Drug projects fail constantly.
>It’s hard, sometimes, for people who’ve worked in other industries to appreciate this. Drug development is a unique combination of very high regulatory burden and very high failure rates, so it’s temping to say that the regulations cause the failures. But that isn’t true. Biology causes the high failure rates – specifically, our lack of understanding of biology.
Calling a field where, after spending tens of millions of dollars of development on a product, only 3.4% of them even work at all a branch of "engineering" is perhaps a novel perspective. If 19 out of 20 bridges promptly collapsed after being constructed, the word "engineering" might have different connotations.
As usual in biotech, when I see someone posting simple, obvious untruths for personal gain, I wonder: delusion, or mendacity? Fast DNA synthesis will not solve the actual bottleneck, which is, and has been for a century, clinical trials in humans.
I see the ‘One True Engineer’ has shown up. Comparing bridge building to engineering antibodies is absurd, a x10^12 error in scale. A better comparison would be to semiconductors, where you will find many ongoing failures, even in current production runs. Also, the mere fact that it is difficult to engineer something doesn’t mean it isn’t ‘engineered’, another absurd contention.
I don’t work in biotech, and also don’t take kindly to being called delusional or a liar. The actual bottleneck isn’t human trials it’s better drugs, and even if trials were the bottleneck, why have you decided that bioengineering needs to be able to design a drug with guaranteed > X% chance of working before it gets to be called engineering? Also, you are ignoring all the other industrial, environmental and agricultural applications of synbio.
Biotech venture has actually outperformed tech venture in recent years. But the venture firms that have done well are specialist venture funds that only do biotech.
All the $100M+ funds on this list are biotech funds: The column group, flagship, orbimed and foresite
Tech VCs have seen these returns and want in, but they have not been major players to date. In 2018 less than 10% of series a rounds in biopharma were led by tech VCs [0]
These days biotech startups can get very big very fast on comparable amounts as tech startups. But he structure in returns is such that you need more concentrated portfolios and low loss rates to do well
I think the narrative about biotech becoming more like tech is a way for tech investors to try to fit more power law driven portfolio strategy into biotech. Bioengineering advances have been incredibly important the last few years, but it is not really correct to compare the pace of product development and value creation in bioengineering to software engineering. I also think it's a bit of a red herring. Biotech VC has done fine on its own. If anything software VCs should take some tactics from biotech funds
I'd be careful reading too much into data on top performing funds. The law of large numbers makes it so that smaller funds are more likey to outperform (and underperform) since they have higher variance. Couple that with the strong correlation that speciality funds are more likely to be smaller and it's not surprising.
A more accurate analysis would involve looking at all VCs and seeing if you can apply those rules in advance:
What would your IRR be if you invested in all the specialty biotech funds as a basket?
The small funds on that list are all tech funds. The larger funds are all biotech.
Biotech funds are "specialist" but they aren't really small. Biopharma is a top 5 subsector of VC (by some estimates it is the 2nd biggest sector after software) and the largest sector by far of healthcare VC, with $17B+ invested in 2018. Orbimed manages $10B+ (though some of that is public equity). Flagship manages many billions (latest fund was $800M+). Foresite manages over $2B, last fund was $668M. Many other examples
I mentioned i was writing a longer post on the topic, that chart is a snapshot. If you invested in biotech VC as a basket recently it would outperform tech [0]. There's more recent data as well but i dont have the link offhand
I look forward to seeing how the newer larger biotech funds do/seeing more recent data during a time where public tech outperformed rest of public market.
Sure, biology might be profitable investment for those who know it well but I think the important part of the GP is that biology doesn't have the potential to "explode" in the fashion that chips exploded with Moore's law.
And the thing is that it's easy to imagine such potential, it's easy to imagine that discoveries will accumulate sufficiently in biology that the field passes from being about discovery to being about engineering, which discoveries can be pumped out easily instead of being hard-won, limited and preliminary. But biology has so much variation and complexity and ad-hocery "all the way down" that nothing is ever as straightforward as you'd want it, nearly everything is done by hand even today and for a reason.
But it's only people who live (and think) in clean room environs like computing that imagine biology is amenable to engineering. It's really not.
Unlike every kind of engineering and physical science, biology is wildly more complex and modulated by almost innumerable variables that are interdependent. I work in a large pharma, and one of the comments I hear often is "I'm amazed that any drug actually works as intended".
With 90% of validated compounds still failing after introduction into the human body (and the thousands of candidates new molecular entity candidate compounds that failed in vitro before that), the track record of human biology to engineered solutions is extremely poor and likely to remain so indefinitely. Just because new techniques are arising to manipulate the assembly language of the body does not mean they will have any better success in setting the right dials to the right settings among the myriad invisible gotchas of disease that inhabit that black box we call 'me'.
Until we can better know what's actually going on inside biology's many black box(es), no amount of engineering, no matter how precise, will reliably (and profitably) improve health outcomes. You can't engineer systems that you don't understand.
But it's only people who live (and think) in clean room environs like computing that imagine biology is amenable to engineering. It's really not.
It's some people used to computing, it's some investors with a streak of optimism and it's some amateur pundits. Humans think in metaphors and "a machine" is one metaphor often used for life. It's not that good but the problem is pure vitalism is often the alternative and that's not good either.
Yep true. But you also missed out that biotechs are trying to cure/treat serious diseases like cancer etc and not simply trying to sell more advertising.
This will be an unpopular opinion, but do you think that money could be better spent on other health initiatives like prevention? Is spending 10-100k per month so rich people in developed countries can live a few months longer really that much more moral?
Consider that orphan diseases are often not worked on due to either 1) not enough rich people have the disease or 2) only poor people have the disease so it's not worth it.
People in developing countries have serious diseases, but there's no money in it so no one is working on it (minus some efforts from people like Bill Gates).
I'd be really disappointed if your opinion is unpopular. I completely agree and find it depressing to see how our culture admires Gates as an idol of altruism. He's had impact, sure, in the same way that someone with an infinite river has the ability to extinguish forest fires but extinguishes a few brush fires instead... just enough to outpace his competitors and lock in the perception of being most benevolent.
For lung cancer - yep - banning smoking would not only massively reduce the rates, but also other lung diseases - China is heading for an absolute epidermic here.
However, for those people that have or are going to get lung cancer or diabetes or other generally ( not always ) 'self-inflicted' diseases - should we just say - sorry - no treatment - it's you're own silly fault?
For heart disease - actually the best treatments are effectively preventative already - in cancer HPV vaccine is aimed at reducing cervical cancer - I've worked with the inventors of that.
In terms of 'whataboutery' of should you be spending money on rich peoples diseases ( whether self inflicted or not ), when you could save more lives in other countries with basic sanitation.
Well that's a whole different question - a question for collective action through governments or charity - not a question for commercial companies. Commercial companies have to find people able to pay for what they do ( which could be governments or charities ).
Personally I don't see why it needs to be either/or - ie why we can't we work on both cure's for cancer and provide clean water?
It's actually very profitable and only overhyped by people that aren't currently profiting from it.
Returns to venture investments in biotech have outperformed tech in recent years, some new drugs are generating $5-10B within a few years of launch, and biopharma is the second biggest subsector of VC after software
Most people haven't heard of the top biotech VCs and companies because they don't blog or do podcasts but they are very good at what they do
And as another commenter said, some of these companies develop drugs that give people with cancer or infants with life threatening genetic disease many more years of healthy life and turn these conditions from death sentences to manageable chronic diseases
Order / disorder transitions constitute a double edged sword. Indeed proteins, dna, rna possess crystalline structure. It may even require decades of microgravity incubation to unlock these oases of stability within the seas of chaos.
The great driver in Biotech is that results are immediately tangible. When we achieve the result that a 100 year old human preserves full memory recall and the neuroplasticity of a teenager. That will really be an accomplishment. One worth diverting even a small fraction of the current trillions spent on military hardware to hard tech bio R & D.
Veeva raised $4M in VC to build CRM for life sciences and currently has a $21B market cap. It's one of the 15 or so best exits of the last decade.
I agree that comparing biology today to transistors in the 1950s might seem hubristic. But even if you don't buy that, I would bet on the notion that there are pretty massive opportunities at the intersection of tech and bio.
I work in Boston and there's tremendous activity in the life science space, as traditionally defined. Doesn't seem crazy to believe that market could support many billion-dollar companies that build picks and shovels.
Yup, and this piece had nothing more than the same fluff we have been hearing the past 15 yrs about bio"tech". Until we can literally simulate a biological system in software (then discovering a new drug would be more deterministic), it will always be a discovery discipline, not an engineering one.
Have you seen the magic that protein engineers have produced? GCaMP6f for instance? You’re right in that it’s still discovery science and is hard but we are making large strides towards proper engineering in certain domains.
I agree with you. I don't think you can learn anything without messing with it and it's in our best interest to learn and mess with it. On the other hand I most certainly would not want to see the SV mentality applied to a field that could bring about a global pandemic.
Oh hell no! I used to joke here that people will eat their hats for the "move fast and break things" ethos when somebody starts Uber for Biotech, complete with sociopathic management and utter disregard for law and safety.
I'm only saying that we absolutely do need to poke complex systems to understand them, but we should do so responsibly. As the reply to the "playing god" saying goes: "we are as gods, we might as well get good at it".
Theranos mostly fooled investors. I worry about companies that mess with the rest of society. "Uber for Biotech" is essentially a recipe for huge loss of life.
Ha, didn't think of that when I wrote it, that's a good point. I looked at it from a Talebian perspective, so to say, and forgot about that approach.
In any case, I stand with what I meant: that complex systems should be treated with the highest degree of care, and with a "not-do-anything-by-default" mindset. The ramifications of interacting with it can, and surely will, be unpredictable.
I think there is messing to find out how it works (works best with man-made technology), and there's applying/marketing stuff that we don't yet understand.
At any. To verify your models about a system, you need to prod it and see if it responds the way your model predicted. Passive observations alone usually aren't enough to even build a correct model. Whether you're interested in understanding low level mechanics or high-level system behavior, the procedure is the same.
Imagine a solar-powered, self-replicating, self-repairing machine that can do carbon capture, clean the air, and improve biodiversity in one go and it's ridiculously cheap and easy to operate and maintain.
Yep a tree.
It always amazes me that people get excited about a new car, new phone or a new software release from a technical angle ( not what you can do with it - that's fine ) - but also don't see the amazing machines called life all around us as interesting at all!
You are always the most complex type of machine in the room - by many orders of magnitude.
Yep, the risks are higher in biotech, yep the rewards are perhaps not as good as a Google etc, and it's way harder. On the other hand it's more interesting and in the end these companies are creating new treatments for cancer or heart failure etc.
Disclosure: I was part of a biotech 'unicorn'.
In terms of whether we are at the cusp of a biotech revolution - I'd say it really kicked off 20-30 years ago and is already here - the new wave of immuno-oncology drugs transforming cancer treatment are a good example.
I think the role of computers in the future is a little overhyped - it's necessary, but not the fundamental driver - the drivers are the new experimental technologies generating the data at unprecedented scale.
I think the tree example is silly, trees aren't good tech because they're not very controllable and often very difficult to modify. And like many biological things, they're not really optimized for our convenience.
Imagine a disease/bug-prone, space inefficient, waste dropping, telemetry-less, allergy-inducing machine that requires continuous water deliveries and a multi-year build process. Yep a tree.
I won't try to tell you that you can't find trees and other biological constructs more interesting, and they certainly pose harder problems, but I'd think (and hope) that the future isn't overly complicated biotech adapted from even more complicated and wasteful natural systems. Right now we have to deal with heart failure and cancer because our crappy bodies are the best we've got, but I would think that the long-term future belongs to engineered systems that we can control, optimize, and understand.
I'm saying it's an amazing machine rather than the perfect machine for a particular task. Ever wondered how it works? It's much more complex that it appears - does the energy capture use quantum phenomena? Do tree's communicate via via fungal network intermediaries?
Do you realize that they grow out of the air - ie all that bulk comes from captured atoms from the air, not stuff pulled in through the roots.
In terms of our crappy bodies - they are incredibly complex machines trillions of cells, each one unbelievably complex - it's astonishing they work at all.
But you are right eventually they fail - and natures answer is to simply reboot ( build a new body from scratch ) - the consciousness you are so keen to protect is simply RAM state that get's blitzed on reboot, while the cycle of life goes on....
We are made from mostly ( 99.85 % ) 11 elements - some gases, carbon, bit of metal
Not saying we shouldn't make stuff - just that people seem to have a blind spot on what we can learn from nature, and technically how amazing it is.
> I would think that the long-term future belongs to engineered systems that we can control, optimize, and understand.
And if we could understand, optimize and control biology - it would be so much more amazing - cells are nanobots - they can already sense, move, change shape to squeeze through gaps, replicate, signal, repair, kill, and control their environment.
In fact those trillion cells - cooperate to make you - you are a collection of cooperating nanobots. Take a look at some forms of life that take this to an extreme, that live as single cells and then come together to build a much more complex structure - https://en.wikipedia.org/wiki/Dictyostelium_discoideum
Big Hero 6 nanobots - biology - been there - done that.
Agreed that nature has plenty to teach us, where I slightly disagree is that controlling and optimizing biology is the better path to future tech.
Yes, cells are nanobots, but they're nanobots "designed"/evolved by nature with deeply complex goals and incentives often orthogonal to ours. Cells come with a lot of evolutionary baggage and many constraints owing to its extensive optimization.
From my perspective, it's like hijacking a crazy project that doesn't belong to us. It's not clear to me that taming insanely complex cells to serve our goals is better than building traditional engineering up to do the same thing.
I compare this tension to ARM vs x86 for mobile devices. Mobile devices uniquely require extreme power efficiency, and ARM processors are dominant because they satisfy that critical requirement even though they're less capable than other processors. Intel wanted to enter the mobile market and they already own the more capable x86 platform, but its power efficiency is a lot worse. Can Intel retrofit ARM-level power efficiency onto its deeply complex and desktop/server-optimized x86 processors before ARM builds up to x86's level of capability?
Honestly, I don't confidently know the answer and you could be right. But I think it's telling that centuries of medicine and decades of biotech have been a slow grind producing very narrow solutions (eg. invented a chemical cocktail to treat one type of cancer, tweaked a plant to contain more of a certain type of nutrient and be resistant to a handful of common diseases/bugs) while ground-up engineering is seemingly exponential, where today's tiny portable processor is orders of magnitude more advanced than the state of the art only a few decades before.
Yeah, nature and biology has a headstart, but we don't have to spend a thousand years figuring out how to develop an appendix.
I'd agree that we can only guess at the future - I'd just saying there is almost a fetishism for man made technology, and a blind spot for nature as it is pervasive.
You say we have made little progress in biology technology - yes and no.
Yes - there is so much more that could be done.
No - you are again ignoring the stuff in plain sight - the vast majority of our food comes from 'engineered' animals or plans - very little from wholly engineered chemical processes.
Breeding is low tech engineering - but it's still engineering - the productivity of commercial wheat is vastly superior to wild grasses for example.
Or look at the variety of dog breeds - each one developed for a particular purpose - from sheep herding & rabbit hunting to guiding blind people.
Now it's perhaps too easy, and not intellectually completely understood - but if you are focusing on outcomes - it's bloody effective.
The right virus - a tiny thing much smaller than a single cell - could wipe out a huge proportion of the population - it's happened in the past.
You can imagine a future where Iron Man fights for human survival with rockets, missiles and high tech, but you ignore biology at your peril.
Good tech isn't just stuff that's subject to control and modification. Look how many species you can choose from. Look at all the stuff around you made out of wood.
Trees create habitat, buffer wind, provide shade, retain soil, and on and on. They're good at what they do.
And every species I can choose from has a seemingly random selection of benefits and problems (and all of them have several issues I listed above) with no clear optimization for any particular human use. Wood isn't a bad material on its own, but trees are a bad tech to produce this kind of material (again due to several issues I listed above).
Being good at what it does is somewhat negated by the other problems it creates, and the fact that some problems it solves are only needed for other suboptimal biological constructs in the first place. There are better ways to provide shade. There are better ways buffer wind. It's good at retaining soil and habitat...only for other biological constructs that suffer from similar problems. I'd be disappointed if our vision of the future is akin to really good soil retainers. Future tech shouldn't need soil at all.
I suspect this sort of nature-fawning is some kind of evolutionary bias; we find beauty in natural biology because we evolved to rely on it. Nature can certainly teach us a thing or two in a way that only millions of years of natural selection can, but tech that we control, optimize, and understand (ie. tech we can actually engineer) should eventually beat it in value.
Uh - premise that we control "tech"? Have you looked at how tech is being used in the wild?
Our natural systems on this planet blow our "tech" systems out of the water by orders of magnitude by their order of complexity and their capabilities. To try and trivialize a tree to the listed attributes without looking how it fits into our entire living ecosystem is so incredibly diminutive and a classic simple narrative of humanity thinking we control this planet and that it was built for our benefit. We are merely here at this moment in time, we might not exist in a thousand more years, a million years? Who knows.
I don't really care that natural systems have incredible complexity and capabilities if they aren't understood, optimized, and controlled for our benefit. If we only have a thousand years left, I want to spend those thousand years engineering things for our benefit, not nature's. Perhaps dealing with advanced soil retainers and other natural systems in the next hundred years is a necessary stepping stone, so be it, but I won't pretend like I think it's an ideal vision of the future.
Appreciate your honesty. I think it is important to note that nature's benefit and our benefit overlap significantly. We do damage to nature and in turn do more damage to ourselves and our own future capability. Probably best to work within the complex model that is keeping us alive then ripping apart only to realize we've destroyed our only form of sustenance. Just my cents.
IMO this is easy to explain by people's infatuation with novelty. Say you took that tree, engineered it to grow to 2 feet tall with bio-luminescent leaves and people would be fascinated by it... until that became old-hat, then it'd be ignored again.
Biology has been eating the world for 4 billion years.
MA: "Bio today is where information technology was 50 years ago."
Bio today is still in the pre-transistor era of IT. There is no tool that doubles our capabilities to manipulate molecular processes every 18 months. No, genome sequencing does not count, it is "read only." No, as amazing and as promising as it is, CRISPR does not count. For the present we continue to merely tinker, which is invaluable and necessary to more forward. Maybe once we figure out how to efficiently, reproducibly engineer enzymes, which truly accelerate molecular processes, we can claim that human-directed bio is eating the world. Until then, nature-driven bio is still on top.
"People who excel at software design become convinced that they have a unique ability to understand any kind of system at all, from first principles, without prior training, thanks to their superior powers of analysis. Success in the artificially constructed world of software design promotes a dangerous confidence."
A VC hyping-up its own book & strategy.
Biology is certainly not an engineering discipline. This claim that it is reflects how little the authors know about biological systems and our limited understanding of biological diversity. Reads like lame sell-side broker research. I recall the promises of biotech of the mid 80’s, the promises of bioinformatics etc etc. Same story, different decade.
Feel the same way about reading this. Felt like gross marketing speak - which to me felt like lazy writing.
To the other persons comment (sorry for not replying to your thread) - of course there are smart people at A16z though it is important to read that smart != correct.
Is this the hype train? Yes to a large extent. However some people at a16z do know what they are talking about. Ex: Vijay Pande is a partner there and an (adjunct) professor of Bioengineering at Stanford. He and the team around him do have an educated guess how biotech today is different from 20 years ago.
I’m quite sure that a16z folks are smart, and perhaps the hyping-up does achieve an intended purpose. What bugs me is the non-stop need to feed narratives which dum-down and embellish reality.
This seems horribly naive and dramatically understates just how difficult drug development is. We simply don't understand enough of the complex interactions that happen in a biological organism, and in the end we have to do very expensive trial and error to determine whether any drug or intervention actually works.
Yes to this. At the Very Big Pharma where I work about 20 years ago we took some baby steps toward adopting an engineering approach toward bio-based systems engineering. Systems biology and in silico modeling of disease had long promised to short circuit the protracted process of drug development. And maybe one day that may yet prove true. But after 15 years of building models of disease or therapy based on data gleaned from lab assays and preclinical/ clinical studies, we found that there were still far too many 'degrees of freedom' in the parameter space of interest to convince the biologists that our models could add sufficient real value to their work for our work to continue.
Fortunately, medicinal chemists have fared better in silico, often employing computational techniques to reduce their search space for a compound desired behavior. But I think no one would describe their models of molecular docking / binding as manifestations of engineering or systems biology.
One thing about this that scares me is looking at the negative parallels, black mirror style, between computing and biology.
When I hear "iterative" design I think of shitty car infotainment or router software that only works half the time and never is updated/supported... Except its under my skin and I cannot get it out.
When I hear designer molecules + DNA I think of DRM and a lack of "right to repair" . But now its inside your body and the business owns it.. err you...
I am very bear-ish on biotech. These are not systems we humans have built from the ground up. Computers are built by humans from the ground up: literally every aspect of a computer is deterministic and documented. And yet even computers can throw up surprises to programmers on a pretty regular basis.
The idea that our understanding of molecular biology is as mature and stable as our understanding of the field effect transistor was in the 1950s is frankly laughable. I predict many many investors getting burnt by biotech, as well as some pretty nasty externalities if regulators buy the hype.
I think the biology revolution will look less like the computing (1940s-2010s) or transportation (1890s-1970s) revolutions and more like the chemistry revolution (1800s-1950s). It will be less iterating on one process and more years of seemingly stagnant progress while people tinker with the last big thing until the next big thing roars out of the gates. Iteration will be brief and seemingly unimpressive. The only real progress will be, that which was expensive and hard will become cheap, but unlike compute that won't necessarily be the driver of progress.
For instance early genetic testing revealed a huge number of disorders that could be linked to single genes that changed the game for how a certain sub-population has kids and treatment for a handful of disorders. Then it turned out a lot of stuff was multigenic or only very partially genetic and progress has stalled. As genetic testing has gotten cheaper we have learned more, but actual usable insights are rare.
Or take lab grown insulin from 1980s, it was a huge deal, everyone worked on similar technologies and knock ins and for a handful of disorders it was a complete game changer that petered out medically speaking. It saw a weird second life in farming and GMOs for a time that helped improve processes but didn't really move the needle in a big way. Then as it got cheaper, and new more efficient tools emergedm it reexploded as the field of biologicals, which have been great drugs since the mid-2000s. And as the technology gets really cheap we are seeing products like lab meats, Impossible leading the charge there, emerge.
Now we are on CRISPR, RNA-sequencing, CAR-T, and tissue engineering all of which, when they hit will have massive impact, but then slow down for a bit until the next big thing revs up.
This parallels chemistry where the theories of thermodynamics and gas laws emerged, sped everything forward, then petered out. Then organic chemistry emerged using a lot of the ideas developed by the previous field, sped everything forward for a while then slowed down. Then inorganic and solid state chemistry etc. But there was never a 18 month doubling it was more like a 10 year lull followed by 1000x in 5 years, if you were counting for example, the number of compounds developed. You could smooth the curve and pretend it was like compute but it's not. We'll just look back in 20ish years, or even look back today at the 1950s, and say 'wow there is a lot of stuff in our life that was based on biological research where did that come from'. Just like someone in the 1940s would look back and be like 'wow where did all this plastic/metal fab come from'.
Good points - but the thing that struck me the most with the comparison to chemistry was how many of those new wonder materials turned out to be more dangerous than we originally thought.
The risks with poorly conceived GMO's are so much higher... but fortunately the technology to do it has been out of the reach of all but a few large companies or institutes who have by and large been very careful.
But now the technology is becoming accessible even to individual enthusiasts.....
It's like nuclear technology becoming in reach of the ordinary citizen.
I've been wondering how long it would be until a16z said something along these lines. I agree completely and am blown away by the progress we have made on informational medicines in the past decade. I would highly recommend to you, if you're a tech-focused person, to get into the field now. I feel like the internet era has matured significantly to the point where the frontier aspect of it has been tamed. Now it's time to move into biology. Look for foundational companies that power the biological information revolution.
I wanted to do this for years now. It´s clear to me, that biotech will become a complete game changer. But what can you do, if you know only the computer science part? It seems like you can contribute only peanuts, if you don´t know one of the natural sciences.
We hire data scientists constantly. We also hire along the full-stack because we make all our software in house. There is tons of low-hanging fruit right now in biotech for CS problem solvers.
I currently live in Switzerland and had no problems getting several job offers for consulting, traditional software engineering companies, insurance companies, Fintech companies and so on.
I also applied to some companies that work in biotech related ares, zero response (full-stack job offers). Well, I think I send one application and two inquiries, still no response. That was very irritating, as it did not happen before anywhere else. But all of them wrote that they want a bit of a bio background, so I attributed it to that.
Eh, the bio requirement is weak but I get it. Without the bio background, best would be to know someone who can refer you to the hiring manager (LinkedIn, etc.) or to be super persistent and asking good questions.
Please do. A simple list of things a CS person should know in order to be useful in a biotech context would be a good starter.
I like to start with some books and simple experiments at home, but have no idea where to begin.
In the past, I thought about going back to college. But I wasted enough of my time there and am still burned. Just give me the relevant bits, skip the fat. But I guess you really need some lab experience to really get biotech and that is not something you just get by yourself.
Get into a lab doing primary research as soon as possible. Full stop. It can be a college/university lab, government, foundation, anything.
In my opinion, information medicines is the fantastic convergence of bio and tech, so I would look for biochemistry labs that focus on DNA and RNA.
Once you have a list of local research institutes, I can help you pick a lab to focus on. Then you just hammer them constantly by email and in person asking questions about the work until they hire you to wash the dishes :)
Eh. You can literally do a PhD anywhere (in western countries) and as long as you're good and your skill set is complementary, you will find a good job in industry science. Going to a top-tier university for a PhD may get you a better network, but ultimately when I hire I really don't pay attention to where the degree comes from. Nor do I care about papers beyond did you get your name on (any author position) at least one during your graduate work. The key is complementary and practical skill sets and being able to speak passionately about your work--even if it's arcane or boring.
Ugh. We've heard all of this before for decades. Yes, progress is being rapidly made on understanding biological processes, but with as many steps back as forward, as we realize there are entirely new categories of things that we didn't even know we don't know.
The hubris on display here is typical of tech VCs, who eventually believe their own hype. If it's hard for humans to comprehend, just rub some machine learning on it... ignore that we're finally realizing that machine learning's limitations are significant, but that we can only discover how wrong they can be on results that humans can double-check.
And the naive application of software engineering buzzwords like "modular" and "iterative" to drug development and biological processes is seriously dangerous. You don't want to move fast and break things when we're talking about human health and genetics.
This article is full of misleading claims and overstatements of reality. It's easy to make things sound great when you entirely ignore the reality of existing and potential technology, but of course I don't expect anything different from these folks.
Let's say you're a tech expert, how do you get involved with the biotech revolution. Is there any resources to learn what is currently available, how you can contribute, etc? I think this field will likely yield some of the most impactful tools and innovations we've ever seen.
This is something I have dealt with a few times. There really isn't a good resource to point someone from a math, informatics, physics, etc. to so they can get an understanding of how the field works. People end up recommending books like Alberts or other things that provide lots of pictures and names in a particular, small area of biology but don't give any idea of what the field covers or what thought processes work in it.
The closest thing I have is twenty-ish pages of notes I braindumped for a colleague who had just entered bioinformatics from computer science, but those already assumed lots of conversations and information already in place.
It's made even weirder because biology is three fields inextricably intertwined (genetics, physiology, natural history). For each one you can start at the beginning, but how do you provide some kind of linear path through all three at once?
With age and humility has your view of the field tempered with time from "Fuck you, bioinformatics. Eat shit and die."[1]?
I totally get your point, and it maddens me daily but I take the interesting + meaningful work/money tradeoff and enjoy being one of the best coders in a building (much harder in pure tech)
Oh boy, that rant. Doubtless the most far reaching and influential thing I have ever written. It gets assigned in bioinformatics classes around the world, which puzzles me in the extreme, and I get emails every couple of months from new readers of it asking my position.
My usual response these days is that if it's raising red flags for them, they should explore what troubles them until they have satisfied themselves rather than taking the word of some random guy on the Internet.
From here inside my head, I think my criticisms are still accurate, nor have I seen any actual refutation of them. Lots of ad hominems. Many people not understanding that terms like "computationally difficult" are precise terms of art, and have nothing to do with intellectual or conceptual difficulty.
With age and humility and, more importantly, absence, my attitude is probably better described as "meh."
If anyone is interested in a revolutionary new type of medical records system where you can copy/paste your entire medical records in a safe way, we are working on a system here called Pau (https://github.com/treenotation/pau).
Imagine if you could just copy/paste your medical records in an email to change doctors. Or participate in a new medical study via a simple copy/paste. Or if you were in an acute care setting and your healthcare providers could instantly call up all your relevant medical history. We are going to enable those sort of things.
Shouldn't we (at least in the US) be more concerned with getting health care costs under control and universal? This manifesto is mostly about making money rather than about general human welfare. The issues are not due to technological limitations.
I mean, yes, we should do that, but we also need medical advances. Right now, if you get Alzheimer's disease, it doesn't matter how much money you have: you will suffer a long cognitive decline until you are no longer recognizable, at which point you will die. The entire process places an enormous social and financial burden on society. We need to fix that.
Personally, I see no greater advance for humanity than curing diseases. Some of these illnesses have been with us since before humans became modern humans, and now for the first time ever, we have a chance at curing or preventing them. I'd take new disease cures over colonizing Mars.
But more to your point, I understand health care costs are crazy (especially in the US), but I am concerned that capping health care expenditures will dis-incentivize companies from taking risks on new vaccines and therapies.
>We are at the beginning of a new era, where biology has shifted from empirical science to an engineering discipline.
What the hell do they think doctors and other healthcare providers have done for all of history? What do they think surgery is? This is such fluffy hype, I feel insulted.
Immune related diseases won’t really be solved unless we really understand how the immune system works. Because of the complexity there seem to be little incentive to fix root causes rather than just inhibit the immunes response I.e. patching
> Bio today is where information technology was 50 years ago: on the precipice of touching all of our lives.
Bio has been touching all of our lives for a long time now, from vaccines to food to basic research yielding large advancements in medical understanding.
What we are on the precipice of may be personalized or customized biochemistry, which is something we desperately need if we ever want to attack diseases like cancer, enable radical regenerative intervention, or tackle the effects of aging.
On a side note, I feel that we should put a moratorium on the phrase "X is eating the world". Right now we're using it whenever something needs to be hyped and it's becoming quite cringy.
When a chimp uses a tool, like a long twig, to pull ants out of the ground, it has used basic engineering (tools) to solve a problem. The species advances because of this.
Engineering is the opposite of entropy. It tries to create order, simple order that can be understood by human brains, so that we can build reliable systems that do useful things. Much of engineering is about making the physical structure (or software structure) of the system have a low entropy (low complexity, no tendancy to rust/degrade over time, statistically very high probability of being able to complete a cycle, etc).
When you build something, taking raw materials from their current state and assembling them into something like a car or rocket ship, is that the opposite of entropy?
not really. what you're doing is using an external energy source to locally order things at the expense of increasing global disorder (ultimately contributing to the heat death of the universe). some people will use a term https://en.wikipedia.org/wiki/Negentropy to refer to this process.
If obeying thermodynamics count, then it counts as a goal. There's a theory that life (and thus evolution too) exists to better dissipate energy.
"England’s calculations suggested that groups of atoms that are driven by external energy sources can behave differently: They tend to start tapping into those energy sources, aligning and rearranging so as to better absorb the energy and dissipate it as heat...
England sees life, and its extraordinary confluence of form and function, as the ultimate outcome of dissipation-driven adaptation and self-replication."
> Buber's main proposition is that we may address existence in two ways:
> 1. The attitude of the "I" towards an "It", towards an object that is separate in itself, which we either use or experience.
> 2. The attitude of the "I" towards "Thou", in a relationship in which the other is not separated by discrete bounds.
Biology is Thou.
- - - -
When Prof. Michael Levin talks about "What Bodies Think About: Bioelectric Computation Outside the Nervous System" (youtube.com) https://news.ycombinator.com/item?id=18736698 he keeps saying, "What if our technology could do this?"
The answer is, our technology already does "this". We are 4By-old nanotechnology. You've heard of Grey Goo? It turns out that the oceans are already "Blue" Goo.
> Gray goo (also spelled grey goo) is a hypothetical global catastrophic scenario involving molecular nanotechnology in which out-of-control self-replicating machines consume all biomass on Earth while building more of themselves,[1][2] a scenario that has been called ecophagy ("eating the environment", more literally "eating the habitation").[3] The original idea assumed machines were designed to have this capability, while popularizations have assumed that machines might somehow gain this capability by accident.
> Marine viruses, although microscopic and essentially unnoticed by scientists until recently, are the most abundant and diverse biological entities in the ocean. Viruses have an estimated abundance of 10^30 in the ocean, or between [10^6 and 10^11] per millilitre.
> Although marine viruses have only recently been studied extensively, they are already known to hold critical roles in many ecosystem functions and cycles.
We keep making discoveries (like the bacteriophages) that show us that we have no real idea what's going on in the biosphere, which includes our bodies. I think it behooves us to wait a few centuries before we go hog wild on the only known ecosystem in the entire Universe, eh?
Not as broad as the range given in the summary at the top of the link, which says 10^12 to 10^28 FLOPS. The lower end (5e10 operations per Joule) is worse than existing hardware: Google’s first generation TPUs were reported as 40 W for 92 TOPS (8-bit) = 2.3e12 o/J, and Apple’s “Neutral Engine” cores are similar (I can’t find specific values for power use, but that was 5 TOPS (8-bit) and therefore plausibly between 1 and 10 watts).
Jokes about Trump being indistinguishable from a Markov chain trained on his tweets notwithstanding, the difference in natural language comprehension between the smartest and dumbest deciles of humans appears to be quite small compared to be small compared to the difference between AI and the average human.
Biotech's dirty secret is that biology is Not engineering (it is still a discovery science)...but the music is still playing folks, dance on! (Sorry, it's early, I'm still pre caffeinated, and this topic always rankles me).