Unleashing the Power of Synthetic Proteins | Hacker News Reader

Unleashing the Power of Synthetic Proteins

alliance.nautil.us

50 pointsdnetesn9 comments

This is our software to help you build your own novel synthetic proteins if you want to give it a try yourself right now in your browser (built with Go and Ember!):

https://serotiny.bio/

No knowledge of DNA or cloning techniques are needed to use it. Drag and drop protein domains with the functions you want to produce a single protein with those functions. And if you wish, we can help place the order for your design with a DNA synthesizer so that the DNA that encodes your new protein arrive in the mail. Entire companies are founded on single good ideas for how to assemble these protein parts. Come up with your own.

It's also a great place to learn about how others are designing their synthetic proteins. The intent is to build up a plain language app-store for all of these various capabilities, as well as keep track of how the various proteins get used. At the end of the day it's not the DNA that does anything in an organism, it's the proteins that are really the machines of your body. If the DNA is the source code, the proteins are the apps themselves. DNA is cool and philosophical, but proteins are useful.

Here is a list and short description of proteins as they've come up on HN in the past few months. I've been trying to keep a list of 'proteins of interest' that have some social, economic, or medical interest to Hacker News:

https://serotiny.bio/notes/proteins/

How often do fusion proteins fail to work? I see a lot of "just stick this catalytic domain to dCas9" and I am not really sure why that works.

More times than not protein fusions work just fine for the desired definition of 'work'. If you are trying to tune the very last efficiency out of an enzyme, or push one organism's survival efficiency past another's, then yeah smashing domains together is not the way you arrive at an evolutionarily polished machine. However if you need to test or create new functionality it seems to work just fine. Biology is extraordinarily robust. A common technique is to combine these to approaches - build out a rationally designed functional protein by fusion, then run it through a few rounds of directed evolution to have the organisms themselves iron out the kinks. This strategy has been very successful.

Just smashing GFP onto every protein in the yeast proteome resulted in ~80% successfully turning green without noticeably affecting the yeast growth.[1] Check out eCD4-IG - it's four components fused to become an extraordinarily effective anti-HIV drug. It is of a different kind and a different order of effectiveness then the small molecule drugs of the 20th century.[2]

[1] http://www.nature.com/nature/journal/v425/n6959/full/nature0...

[2] https://serotiny.bio/notes/proteins/ecd4ig/

Some context: the author, David Baker, is a preeminent figure in the world of protein structure prediction. His lab has consistently had the best protein structure prediction software out there (Rosetta). They recently released a paper detailing the structures of a large number of unknown proteins using their technique as an aid [1].

Some of the ideas in this article are quite cool, but I think it's important to add a caveat: we don't often know what structure will yield what result. In other words, if I want to do X, I don't know what protein structure will get me X, which makes it hard to engineer such a protein.

Nevertheless, when you have some understanding of the problem, you can in fact engineer proteins. Baker gives several examples in the article. In many of those cases, the best approach is to start with a variety of engineered proteins, then use in-vitro simulated evolution to pick the winners and get pretty substantial improvements over the original design.

Protein engineering is quite a cool area, and hopefully over time we'll get better at understanding the other parts of biochemistry so that we can more effectively design useful proteins.

[1] http://science.sciencemag.org/content/355/6322/294

There are a lot of approaches, and they do not all require a brute force simulation. Nature herself seems to have a library of functional domains from which she can mix and match to produce a larger mechanical device - in addition to the atomic resolution mutations, and tertiary structural features. There are a lot of very useful ways people do design proteins right now that do not involve structural prediction.

Further, very much unlike the digital sciences, a gigantic library of parts, functions, routines and 'example code' already exist. If you want a piece of nano-technology that does X, just go look for an organism that lives in a condition where X is useful - and in there genome it will be. We have a huge pile of hyper-advanced alien source code that we can dig through. We don't have to reinvent it all on our own.

I'm convinced that synthetic proteins are the most viable way to Drexler type nanotech. Biology demonstrates that it works, we "just" have to understand how.

At the sub 10nm scale you need stiff linkages to have predictable and consistent structures. Covalent 'organic' bonds give you stiff linkages at that scale while metals are goopy and electronically inconsistent. Proteins are organic nanotechnology where the 3D feature sizes of the (wet) tools is measured in fractions of angstroms. The best Intel can do is ~ two orders of magnitude larger resolution, 2D, dry.

Could somebody with more expertise explain how modern day protein folding simulations work? (Like Rosetta) I'm only familiar with the folding@home method of brute force molecular dynamics simulation. What kind of computational power does it take to go from an RNA sequence or polypeptide chain to stable protein with these algorithms?

This has always been a subject that has fascinated me, ever since high school when my biology teacher told us about folding@home. I went into computer hardware, so I always saw this problem as a computational power problem. I always believed that if we could figure this out, we could do a lot of cool things like make nanotubes and cure diseases. I saw it as the key to the nanotechnology revolution.

Rosetta is a software for protein structure prediction, not for protein folding simulation. There are a lot of hipothesis about protein folding, but the mechanism is unknown yet. So simulations using molecular dynamics can help us to understand how proteins fold.

On the other hand, protein structure prediction methods like Rosetta try to create high quality atomic models of proteins without follow a physical model. The goal is a good final model, not the fold event.

Rosetta uses a method of fragment assembly where fragments (few amino acids) of proteins with structure solved experimentally (usually X-ray crystalography) are assembled to create the new structure. The fragments have sequences of residues (amino acids) identicals or similar to the protein being modelled. This is an optimization process and uses methods like Simulated annealing. So, computational power is very important to explore a lot of protein conformations.

Reply on news.ycombinator.com