Launch HN: Radix Labs (S18) – Downloading drugs with programmable lab automation

27 points by dhash ↗ HN
Hi HN, ​ We're Dhash and Ian, the founders of Radix Labs. We build software that lets biologists automate and transparently share their work. Our software glues together biologists' existing lab hardware and provides a seamless way for biologists to execute their protocols on it. Check us out at http://radix.bio/ .

We think lab equipment is like mainframes were in the 1980’s - you have to program each one as a special snowflake. All that we've done is built an operating system so you can write a program once that runs on all of them. Just like developers write apps for android, they write recipes for Radix. In the future with our technology, you’ll be able to download a recipe for making a coca-cola and if you have the right robots and precursors in your lab, they’ll just start making it.

Technically, we do this by making biologists describe their protocol in a formal language with verified operational semantics. We do this with a web UI, so they don’t have to write “real” source code. We compile this protocol to our IR, deriving control and data flow graphs to inform our place-and-route algorithms. After the program is realized to robot/lab instrument commands, we generate code to be run on the automated machinery. If the entire protocol can’t be totally realized by automation, we generate natural language instructions for humans to interact with the runtime system.

We go through this process to gain features you’re used to in your “normal” operating system. We provide a virtual memory abstraction so that programs compose, concurrency primitives for executing tasks in parallel and synchronizing access to fluids, and drivers to support a wide variety of different lab hardware.

These abstractions differentiate us from companies like Transcriptic and Emerald Cloud Labs by allowing us to write programs with branching control flows, support for on-premise deployments, a wider variety of compatible hardware, and an advanced optimizing compiler to allow users to specify optimization criteria like minimum execution time, minimum consumable usage, or max flow to run a lab as a high-throughput assembly line.

Right now we’re compatible with some pipetting robots (Opentrons and some Tecan robots) and we're working on adding more devices. Contact us and let us know what you need automated. Launch HN: Radix Labs (YC S18) - Software for Bio Lab Automation

We’re doing this to bring better lab-scale tooling to biologists, it currently costs 2.3B to get a new drug to market. This is too high to the point of where promising drugs are scrapped for economic reasons, and we think we can help fix that.

We’ll be here to answer any questions you have about us and our product for the next few hours as @Ian_Paul and @Dhash. Also feel free to fill out this (http://radix.bio/contact-us.html) form and we’ll answer any question you have over email

8 comments

[ 0.36 ms ] story [ 31.1 ms ] thread
Congrats on the launch! You guys seem like a very different take on "computational biology". As a heads up, it looks like the twitter link on your site just redirects to the contact form.

Good luck!

Thanks for the suggestion, I'll fix it now.

The compbio folks are cool, our product is written in Scala to easily integrate with their existing pipelines.

What are some concrete synthetic biology workflows whose labor costs have been reduced by this?

Why do you advertise this as "being for biologists" and also "for downloading drugs" at the same time? Biologists do much much more than just synthesize drugs.

How does the program "execute" a decision once, say, it detects that the percent molar composition of molecule A exceeds that of molecule B?

Some concrete workflows are easy things like aliquotting and PCR, and as complex as PACE, a protocol for directed evolution.

The "for biologists" and "downloading drugs" only account for a small part of what we do. We use those phrases to get the point across, but you're totally right, they do a lot of stuff that we can automate such as parameter sweeps over a protocol to empirically increase the yield of the reaction.

We use normal conditionals, so if you can write an if statement with the desired branches, we can execute that. Both paths are statically allocated, currently there is only limited support for deferred routing. It's something i'm working on!

Do people still have to specify position of {fluid 1 in well 2}=>requiring a dependence on the well or does it allow for complete position independence?
No, complete position independence is allowed and encouraged. People can put constraints on well location, like “use this plate”, but general allocation is handled by the compiler. If a well needs to be bearable, for example, that is allocated into a well that satisfies the temperature constraints of the heating operation.
What are some microsensors that can be used to detect concentration of a protein or molecule that can then be used to make a "decision"? E.g. aggregated proteins or concentration of a particular mRNA or 4-HNE or lipofuscin accumulation?
Thanks for making this.

Will people be able to distribute their apps through an app store (a common feature for modern operating systems)?