> Biology usually doesn't work.
No. This is not true. Biology usually works, it's that the most interesting academic things tend to be on the bleeding edge, and that's where things are not robust.
Of course, I'm not saying that there isn't debugging to be done, but it's not THAT hard. I'm doing something relatively difficult literally in my garage right now (growing a strain with a doubling time of about 6 hours). I can reproducibly make the anticancer compound in one of my strains and am working on debugging and getting a second strain running. (https://benchling.com/ityonemo/f/cHjBceoz-project-marilyn/et...)
> Even the simplest wet-lab reactions take very exacting conditions to work.
Not true. For example I have done over 300 gibson assembly reactions and have taught an intern to do this, he made 50 constructs in 2 months, with time left over for him to biochemically test 25 of them. My procedure was literally, take 1 microlitre of each DNA (don't even bother measuring), and throw on top an equal volume of gibson mix, and then go. Worked nearly every time.
It is quite true that equipment requires a high capital expenditure and it's hard to start up, but those problems are able to be overcome. (e.g. http://blog.indysci.org/starting-a-lab-under-budget/) For example, I have bacterial growth incubator that was being thrown out by herbalife when they upgraded their microbiological QC lab.
When the parent says that "even the simplest wet-lab reactions take very exacting conditions to work", they're not really talking about transforming cells with plasmids and autoclaving plates. The conditions are exacting, but the equipment to provide those conditions is not exotic.
The hard, expensive part of biology is the part where someone spends weeks/months/years using those tools to make a new construct that should work, but then doesn't work, because unknown. Then you spend months trying to figure out what factor you're not accounting for, then you give up and start over from a new angle. Then you repeat for 5-7 years, and someone gives you a doctorate.
Saying that biology mostly works is like saying that self-driving cars are easy because we can reliably attach computers to the steering wheel and the gas pedal. It's the parts we don't know about that cost the money.
I mean, I'm not a complete idiot here. I do have a PhD in biophysics/chemistry/biology and four years of postdoc experience. I've designed expression constructs nearly from whole cloth; I've done spectroscopy on metastable peptides; I've produced, characterized and sent for testing a mutasynthesis product; I've improved an enzyme four-fold. Of course all of those things were done in 'real labs', but the last two were done under budgetary pressure where there was a strong emphasis on getting things done cheaply (UMD is not very well-funded, J. Craig Venter Institute was making a painful transition from hard money to soft money). A lot of those lessons I learned under budgetary pressure translate to doing things in the garage, except, more extreme.
I think the primary obstacle to people doing garage biology is not that the biology is hard, but that there are generally two classes of operators, those who have little biology experience, and those who have worked under coddled environments. The real trick is to find things that are maybe a little hard, but not insurmountably so, but with one or two difficult pain points. And which are also interesting. The first class of operator is not generally capable of identifying those qualities, the second class of operator is out of the habit of identifying those qualities.
Biology really does mostly work, you're just looking at a biased system where people are just throwing money at problems and also looking for reasons to justify continuing to throw money at them (so they have to do hard things). That also creates silly folkways like the 'biology is hard' trope which makes the operators miss obvious things because it's just easier to say 'biology is hard' than to think about the problem and figure out how to solve it cheaply and effectively.
In case you're wondering, my definition of "hard" is the CS definition: there are "trivial" problems ("send this structured text over a network connection"), there are "straightforward" problems ("scale this website to X requests per second with a median latency of Y"), there are the "hard" problems ("translate english text to another language"), and there are the problems we suspect are intractable ("solve the halting problem"). Pretty much all of the wealth generation in software over the last 30 years has come from the "trivial" and "straightforward" problem sectors. Even Google, when you back out a bit, was the result of the careful solution of straightforward problems. It's not like they "solved" information retrieval or machine translation -- they just took existing research, and scaled it.
In contrast to CS, biology is loaded with stuff that shows potential, but that we just don't understand very well. The money isn't in making bacteria glow, it's in making bacteria glow in a way that provides a reliable biosensor, or making those bacteria produce things that are safe to put inside people. The limiting step here isn't access to robots -- it's our horrible understanding of the underlying biological systems. There's very little (if any) biotech stuff out there that is a "straightforward" problem away from commercial success.
And, while the design cycle for biology is longer than CS, it's not that bad, especially now. Given that I have decades of computer programming and wet biology in my background, I guess you can 'trust' that I'm not one of these guys that thinks progamming a cell is as easy as programming a computer. But I do also frequently lament that the approach to biology is often the 'script kiddie' approach.
Among academics and in industry, I OFTEN see that the operator's understanding lacks depth (not the field). We were having a problem with DNA shearing once, and I asked all of my coworkers if they could explain to me the chemical mechanism of DNA shearing and not a single person could even postulate a testable experiment that could have a positive effect on both what we were trying to do and our global understanding.
In another case, I asked around if anyone knew the mechanism of electroporative transformation into E. coli. Looking it up (the answer had been known for about two decades) resulted in a small change in our transformation procedure (at no extra cost!) and a reproducible 20x improvement in the transformation efficiency of 500+kbp DNA.
Yeah, maybe "we just don't understand it well", but it's often because the scientists don't understand it well, not the corpus of humanity as a whole. Now: I'm not saying we have a comprehensive understanding of biology, I'm saying it what we the culture have is enough that simple, useful, garage operations are really in striking range.
Yeah, I know. I was (unclearly) trying to draw the comparison to the research into correctness. It's impossible for the general case, but people still try to make useful progress with the specific cases.
"A large reason why these have not been commercialized is because of the strong IP culture that permeates biology, and the chilling effect that has had."
I sort-of agree with this. I think one major difference between biology and CS is that CS has a long history of public knowledge about systems that tends to build on in ways that become more abstract (and more powerful) over time. A lot of biological knowledge is public, too, but most of the commercially relevant knowledge is locked up.
"This also creates a bias towards intractable problems. The more difficult your problem is, the less likely it is to have been patented, so there is an incentive to work on those problems."
I totally disagree with this. I've never encountered anyone in the research world who has shied away from a problem simply because there were patents involved. There are a lot of patents in software too, and in my experience, we treat them the same way: we ignore them, and deal with the consequences later.
"Among academics and in industry, I OFTEN see that the operator's understanding lacks depth"
Yeah, OK. I grant you this (I've encountered it, too), but I still don't see how this is meaningfully affected by the stuff we're talking about. Bio hacker-spaces and robot-rental startups do not make this problem go away.
"I'm not saying we have a comprehensive understanding of biology, I'm saying it what we the culture have is enough that simple, useful, garage operations are really in striking range."
Yeah, well...I'm still waiting for examples. ;-)
I tend to believe that the market for good biotech ideas is efficient: there are enough extremely talented people working in the space, and enough money chasing plausible outcomes, that if there's an idea out there that doesn't take decades of basic science to find product-market fit, it has a company working on it already.
More to the point, though, I definitely do not believe that the rate-limiting factor for the creation of new biotech startups is access to machinery. It's access to knowledge, and therefore, this kind of "we're on the cusp of a biotech revolution!" stuff is, well...hyperventilation.
That's great for your cell-based experiments; and many of the greatest biology discoveries were made with this toolset in the first half of the 20th century.
But what about in vivo experiments with mice and rats? What about X-ray crystallography? What about high throughput screening and counterscreening in the range of millions done before a compound is even considered to be a drug candidate?
This is hard stuff, and it's necessary stuff. Exacting conditions are absolutely essential to achieve the highest chance of reproducibility. Even so, it has been famously reported that somewhere around 11% of published findings can be reproduced independently[0].
[0] http://www.nature.com/nature/journal/v483/n7391/full/483531a...
The expensive stuff is when you go into humans.
I can only guess what "five arms" you are referring to, but toxicity studies and efficacy studies often require different animals. Not just mice but transgenic mice, or nude mice with xenografts, etc. This has to be pricey to get the statistical power you need for a drug candidate.
I've also heard that the FDA likes to see more than one animal, so either rats or monkeys may be dosed as well. It all adds up.
Five arms, refers to an experiment with three dosages, a null control and a positive control.
Yes, you can outsource things, which is exactly what I'm doing (it's cheaper and more ethical), but also the model has become really good where you don't have to sack an animal for each time point. That cuts down on costs by a lot.
They started getting reproducible results, then, major improvements, then the facility got shut down for political reasons and she lost her job.
No, not hard, just crazy expensive.
>Biology usually works
Under the most basic circumstances. Most startups won't have a business model based on doing PCRs with a master-mix.
>For example I have done over 300 gibson assembly reactions and have taught an intern to do this, he made 50 constructs in 2 months, with time left over for him to biochemically test 25 of them.
Likely at a corporation with money to blow on facilities, equipment, and reagents.
We are talking bare-bones start-up here.
>I'm doing something relatively difficult literally in my garage right now
If you want to make impure vitamins and cram them into pills, yeah I guess, but if we are talking even rudimentary biotech level work, then a garage just won't cut it.
I think of siRNA and its revolution in biology -- it went from discovery to practical use everywhere in record time, just like CRISPR -- but the direction of research itself did not take that same radical shift. And medicine was improved too, but incrementally.
1. Biology usually doesn't work. Even the simplest wet-lab reactions take very exacting conditions to work. This requires precise, expensive, equipment and endless optimization of protocols. The result is a tremendous drain on resources to get what usually amounts to a negative result
2. Bio is inherently perishable. You can't let a start-up project linger as you chip away at it for 3 years. Reagents expire, lab-cultures will mutate, equipment rusts/breakdowns/gets contaminated for stupid unforeseeable reasons.
3. There is a stupidly large legal burden if you ever want to commercialize. At every step of the process the USDA, EPA, FDA, and patent office have rules and regulations to slow you down. It might take decades to take a product from the lab to the store self, if it gets there at all.
Bio isn't computing. As a geneticist, I've seen it a hundred times where some start-up know-it-all walks into the biotech field and thinks its "just like software". It isn't, and never will be.