I had the very great luck to study control systems as an undergraduate with James Collins, one of the people responsible for the first big "genetic toggle switch" results in 2000 : https://www.nature.com/articles/35002131
The authors of this piece note:
Yet despite our progress so far, genetic circuit design has often been characterized by a manual and failure-prone process. Engineers often spend years creating a functional design through trial-and-error.
This is still absolutely true.
Part of the challenge of the synbio field is to do better than this while still being on the sane side of the "andy grove fallacy", which is loosely stated as "hey biologists, transistors are easy now; what's the holdup?" . The holdup, of course, is that 4.5 billion years of monkeypatching makes for a really hard slog.
I've never heard of the Andy Grove fallacy, but it is extremely relevant. As a bioinformatician actively working in the field, I get the same statements from people in the tech industry. Often it comes down to: "why don't you just apply this cool new technique and all your problems will be solved".
I think people need to realize that biology and tech are disciplines that might traverse on the same track, but in opposite directions.
Biology is going from top-down where we have an incredibly complicated system that we need to dissect and reduce down to workable fundamental parts. In tech/computer-science, we are going bottom-up where have man-made fundamental parts that we build up into incredibly complicated systems to perform a function.
We are kinda at the point where both disciplines are starting to intersect each other on the same track; hence the popularity of biotech right now.
Biology is going from top-down where we have an incredibly complicated system that we need to dissect and reduce down to workable fundamental parts.
The additional problem there is biology (and the underlying physics) is currently in a state where the known-unknowns continue to increase at an exponential rate. Exploration to reduce physical/biological systems into workable parts continue to reveal even further constituent sub-systems, which in turn opens up entire new fields of research and new unknowns to be explored (seemingly ad infinitum).
Yeah agreed. I don't think most people actually understand the level of complexity involved in biology. A cell is literally a bag of molecules where everything inside is just bouncing around. And somehow, things self organize and stuff works and performs functions. And that's just one single cell. My background is in developmental biology, we have some mechanistic answers to how cells organize to maintain tissue polarity, but nothing really concrete.
One thing I discovered recently is that genes form a 'gene regulatory network' that bears some analogy to neural networks. Genes are like neurons, each with inputs and outputs interconnected in a complex way between them, and the chemical messages they pass around are proteins. So there is a whole 'neural net' inside each cell - this information put cell complexity in perspective for me.
I think about this sort of thing a lot, especially in regards to software. What is one's level of understanding of a system? Of the general architectural principles of similar systems in general? In software we still have incredibly crude tools to describe architecture which typically gloss over almost all relevant details. Between the raw source code and your typical multi-layer diagram there is a gulf of understanding and missed opportunity. Biology has an even worse problem, with even more complicated machinery that is even more difficult to understand (you can't simply step through code in a debugger to see what it's doing, and the "code" was evolved so you don't even really know what the operations are (such as gene expression and suppression)).
After listening to one of the world's top organic chemists, one realises that humanity is still in the stone age when it comes to industrial and technological systems when we look at what is going on inside a single biological cell.
This includes chemical transport, energy transfer, system construction, chemical processing and purification, adverse chemical reduction and removal, etc.
The more I look at the information obtained about biological cells and their operation the more it becomes obvious that certain popular models related to biological systems are in the category of "the emperor has no clothes" ideas.
Biological systems are so complex and so integrated and so coordinated that we still have so little actual knowledge on how and why they work the way they do. So duplicating the various industrial processes that occur within a cell with any finesse is still a long, long way off.
What you're seeing here (and in other places), is us advancing to the tool-making stage of biological nanotechnologies. And this is one aspect of it. We've been gorging on grounded knowledge in the past 2 decades with the ability to readily read DNA, and we're now at the stage where there are enough robust tools in enough different biological disciplines that we no longer just 'find' tools, but we can actually start to create them.
Between companies providing the ability to cheaply synthesize DNA in bulk (Twist), companies essentially doing machine-guided evolution on bacteria (Zymergen), synthetic therapeutic circuitry (Cell Design Labs), and a host of brand new entrants to the synthetic biology space, we've a lot more capability than we did just a few years ago.
We've found enough stones that we're now starting to build our own primitive (3D, atomically precise, wet, self-replicating nanotechnology) tools. CARs, SynNotch, HBB[T87Q] are all creative upgrades to existing biological systems used where our own bodies have failed, and all only very recently effectively delivered. And we can achieve those pragmatic wins without having a complete and thorough understanding of the entire system.
There is room for multiple approaches - as biology itself proves. Asimov has a pretty interesting approach and a really good team.
Tooling in the regulatory space of biology allows one to subtlety alter the mass transfer within existing networks - which can be harder to figure out, but often allows a delicate touch to cause a strong effect. And, especially, if you want to work in non-mammalian systems, I suspect you can actually start to build otherwise orthogonal networks.
At Serotiny we are playing with designing proteins and synthetic receptors whole cloth - which, though it can generate entirely new functionality, also brings with it all the challenges of dropping an invented capability into an existing system. You can be very creative, but integrating that creativity can itself become a challenge downstream.
Honestly, I think they are complementary approaches, and in the end, I suspect many of these approaches will ultimately converge (again, see 'life'). But we require expertise in each area first.
You can model gene regulatory networks at a high level with recurrent neural networks. It's interesting to think that nature re-invented the same computational principles with different implementation details.
16 comments
[ 4.1 ms ] story [ 49.0 ms ] thread[1] http://igem.org/
The authors of this piece note:
Yet despite our progress so far, genetic circuit design has often been characterized by a manual and failure-prone process. Engineers often spend years creating a functional design through trial-and-error.
This is still absolutely true.
Part of the challenge of the synbio field is to do better than this while still being on the sane side of the "andy grove fallacy", which is loosely stated as "hey biologists, transistors are easy now; what's the holdup?" . The holdup, of course, is that 4.5 billion years of monkeypatching makes for a really hard slog.
I think people need to realize that biology and tech are disciplines that might traverse on the same track, but in opposite directions.
Biology is going from top-down where we have an incredibly complicated system that we need to dissect and reduce down to workable fundamental parts. In tech/computer-science, we are going bottom-up where have man-made fundamental parts that we build up into incredibly complicated systems to perform a function.
We are kinda at the point where both disciplines are starting to intersect each other on the same track; hence the popularity of biotech right now.
The additional problem there is biology (and the underlying physics) is currently in a state where the known-unknowns continue to increase at an exponential rate. Exploration to reduce physical/biological systems into workable parts continue to reveal even further constituent sub-systems, which in turn opens up entire new fields of research and new unknowns to be explored (seemingly ad infinitum).
https://en.wikipedia.org/wiki/Gene_regulatory_network
I think about this sort of thing a lot, especially in regards to software. What is one's level of understanding of a system? Of the general architectural principles of similar systems in general? In software we still have incredibly crude tools to describe architecture which typically gloss over almost all relevant details. Between the raw source code and your typical multi-layer diagram there is a gulf of understanding and missed opportunity. Biology has an even worse problem, with even more complicated machinery that is even more difficult to understand (you can't simply step through code in a debugger to see what it's doing, and the "code" was evolved so you don't even really know what the operations are (such as gene expression and suppression)).
This includes chemical transport, energy transfer, system construction, chemical processing and purification, adverse chemical reduction and removal, etc.
The more I look at the information obtained about biological cells and their operation the more it becomes obvious that certain popular models related to biological systems are in the category of "the emperor has no clothes" ideas.
Biological systems are so complex and so integrated and so coordinated that we still have so little actual knowledge on how and why they work the way they do. So duplicating the various industrial processes that occur within a cell with any finesse is still a long, long way off.
Between companies providing the ability to cheaply synthesize DNA in bulk (Twist), companies essentially doing machine-guided evolution on bacteria (Zymergen), synthetic therapeutic circuitry (Cell Design Labs), and a host of brand new entrants to the synthetic biology space, we've a lot more capability than we did just a few years ago.
We've found enough stones that we're now starting to build our own primitive (3D, atomically precise, wet, self-replicating nanotechnology) tools. CARs, SynNotch, HBB[T87Q] are all creative upgrades to existing biological systems used where our own bodies have failed, and all only very recently effectively delivered. And we can achieve those pragmatic wins without having a complete and thorough understanding of the entire system.
http://www.wired.co.uk/article/at-home-with-the-dna-hackers
As a cell biologist turned programmer, i am certainly interested to see if anyone can make something really work in this domain.
But also a bit frustrated that people are focusing on genetic circuits, which are slow and boring compared to protein-based signalling networks!
Tooling in the regulatory space of biology allows one to subtlety alter the mass transfer within existing networks - which can be harder to figure out, but often allows a delicate touch to cause a strong effect. And, especially, if you want to work in non-mammalian systems, I suspect you can actually start to build otherwise orthogonal networks.
At Serotiny we are playing with designing proteins and synthetic receptors whole cloth - which, though it can generate entirely new functionality, also brings with it all the challenges of dropping an invented capability into an existing system. You can be very creative, but integrating that creativity can itself become a challenge downstream.
Honestly, I think they are complementary approaches, and in the end, I suspect many of these approaches will ultimately converge (again, see 'life'). But we require expertise in each area first.