News for hackers. Stuff that matters.
Re:Transcriptic for YC biotech startups by dnautics on Monday December 08, 2014 @03:29PM (#8718961)
The biggest pain point I see in this market is that experimental parallelization is an intervention-heavy process. Ignoring the equipment costs, it's also capital-intensive (unlike say deploying to AWS). And finally, obtaining usable data still requires experiential knowledge. There's something about knowing and feeling the data (yeah, that's awfully fuzzy) that is still an important part about obtaining good results [0]. So for any biologic process that is parallelizing the operators are going to want to own the machines anyways.

In order to capture a real market, you're going to have to figure out a way to offer parallelization services - be given a non-parallel experiment with certain parameters and scale it up on behalf of the users. So, the user has an experimental plan and just 'hands it over' to transcriptic. I still worry about the experiential knowledge part, putting the experimenter one step away from the experiment is potentially counterproductive.

[0]http://onlinelibrary.wiley.com/doi/10.1002/pro.2339/full In this paper, the grad student (and lead author) who had spent four years of her graduate work on a previous paper had a nagging feeling that the data were strange. By actually looking at the wells, she figured out post-publication (with nothing to gain) that the protein was sticking to sides of the 96-well plate and causing the observational data to be artefactual. Then there was the question of how to do more experiments to prove that was going on. And then the political problem of convincing her grad advisor to publish a retraction (well at least it was a retraction worth a 10 page paper and a new citation. The story has a happy ending; she got a position at a pharma company largely on the back of her due diligence).

Re:Transcriptic for YC biotech startups by frisco on Monday December 08, 2014 @04:32PM (#8719331)
I think the opposite is actually true: you get better data when you don't feel the samples. This is a highly unpopular view right now but I have to wonder what's going on when things like Amgen getting 11% reproducibility of foundational papers (http://www.nature.com/nature/journal/v483/n7391/full/483531a...) are happening. There are things that are hard to automate because of their mechanical nature, but I think that for the reproducibility of science somewhat distancing the human from the process is a good thing. Of course, this adds short-term costs (which may sometimes be unacceptable) and requires a lot of behavior change for how people are used to working. I will say that we put a lot of thought into giving you the fidelity of interaction such that you can still make "breakthroughs from your errors" on Transcriptic, and improving those reporting capabilities are an ever-ongoing process.

I'll also say that this challenge is bigger than just one company. Some things may make more sense to do via Science Exchange, for example if the method requires some very customized hardware or there are only a few experts in the world who are sufficiently familiar with an unusual method's sensitivities. I'm also excited to see what Riffyn comes out with to help labs understand where reproducibility comes from. We're just getting started, but I can't see a path forward that puts more humans at benches rather than less. The humans should be free to do real science.

Re:Transcriptic for YC biotech startups by dnautics on Monday December 08, 2014 @05:30PM (#8719665)
I have exactly the opposite opinion. Even scientists are captiviated by scientism - the idea that there is something poisonous about human subjectivity and imprecision and that removing subjectivity (and gathering more data) is necessarily a good. Sydney Brenner, for example has a 'money quote' about the path that biologists take: "low input, high throughput, no output science". Note that this quote doesn't make sense in an environment that doesn't faddishly flock to high througput 'big data' solutions.

Your example is greatly flawed. The biggest consumers of highly parallelizeable workflows is the pharmaceutical industry. Highly parallel medchem was a big fad and the number of drugs that it produced for its efforts is disappointing. The fact that 11% of Amgen's results are irreproducible is if anything a condemnation of parallel scaleup, at least in the context of an operator with a strong motive for selective interpretation.

Another big problem is that when you bring your numbers up, you 'get what you are looking for'. Precision optimization can optimize for an artefact. I joke I like to make is that sloppy science is good, because if you keep seeing the same result under a noisy platform, what you're seeing is probably real and, more encouragingly, robust.

Re:Transcriptic for YC biotech startups by zmanian on Monday December 08, 2014 @05:03PM (#8719520)
You are absolutely correct that the lack of standardization in assay preparation harms productivity and reproducibility in the sciences.

Molecular biology has embraced standardization and automation to a great degree but it is an outlier.

I've encountered enormous resistance from practitioners trying to apply automation into cell biology.

Biology is enormously held back from the tools for improving productivity have to come from computer science but there is huge amounts of interdisciplinary friction.

Re:Transcriptic for YC biotech startups by dnautics on Monday December 08, 2014 @03:35PM (#8718988)
that said, I don't want to be a total negative nancy! There are some services (like flow cytometry) that I think I might use for transcriptics.

I think you could also capture an interesting market if you offered "evolution as a service" using your kit.