Deep Mind AI predicting how proteins curl up from a linear chain of amino acids into 3D shapes that allow them to carry out life’s tasks. DeepMind AI will massively speed up drugs for protein-based problems. Deep Mind revealed enough of their methods for others to make use of them. It will only be a few months before other groups match DeepMind’s success.
Proteins do most of the work inside all cells. Human cells, animal cells, plants cells and all biology. This means besides improvement in human medicine there will also be agricultural and biotechnology advancements.
The functions performed by a protein depend upon its three dimensional structure. Now we will rapidly gain more control of those structures. We have been able to synthesize proteins just as we have been able to synthesize DNA. Now we will be able to create protein sequences that form desired dimensional structures to generate the desired functions.
Having this structural control of proteins will also accelerate the science of understanding the functions of proteins.
UPDATE: This is a big advance in protein folding prediction but a lot of improvement is needed to be able to get to the point where we can predict protein folding enough to speed up drug discovery. There are some claims that we need to go from an average of 0.16 nanometer accurate predictions to an average of 0.03 nanometer accurate predictions.
SOURCES – Science, Deep Mind
Written By Brian Wang, Nextbigfuture.com
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Solved is such a strong word for computations that ignore the presence of water which all proteins sit in. Until someone computes a snowflake which is just water I'm still a skeptic that we're not just playing in a sandbox that is divorced from reality.
True. However there should be multiple possible amino acid chains that should result into a target 3D protein structure (withing a given design tolerance) for your application.
It's amazing practical application 😀 (with caveats)
This solution, allows a user to input an amino acid chain and provides the probable structure of the protein. This is already a BIG deal. Doing this experimentally costs upwards of USD 100K and takes about a year.
Now users can start spamming amino acid sequences into the input and look for promising structures in the output, for some application.
If they get lucky, we could have super materials (think the plastics revolution of the 20th century, but now with proteins in the 21st century). If we get _really_ lucky, we could get a system of proteins that combine to form a useful nanite.
Even if we don't get really lucky… This is just the first solution to cross the 90% GDT metric. In the future there should be other better (more user friendly) solvers that will allow us to specify a desired set of properties and to output the protein structure and amino acid chain, which will accomplish it. At this point we wouldn't need luck to design and build our nanitesG
Layman this for me. What does this mean for us? Is this an academic solution like Fermat's theorem or are there some amazing practical applications here?
The computer they use isn't nearly as powerful as the Cerberas CS-1. It would be interesting to see what protein engineering they could do with a Cerberas CS-1(and the 2.6 trillion transistor chip has been on the test bench for months now)
Perhaps the general approach to the reverse calculation would be "do a lot of them", i.e. create a library based on some patterned set of RNA sequences (progenitors of protein), characterize the results, and apply another level of AI to that. Does sound a "bit" challenging.
On the other hand, maybe the actual real world problems involve getting a shape that will (e.g.)
– fit one "branch" into opening X in cell wall location Y
– while attaching to another molecule Z to some other part
i.e tranport stuff to cells,
and that it will be less important what the rest of the shape will be (hopefully non-toxic, etc.)
Needs to be more jumping up and down excitement about this. I concur with Dr. Pat that this doesn't get you from "need this protein shape" to "start with this RNA strand" however it's a major step forward.
This achievement is to go from protein structure -> folded shape.
You seem to be talking about going from desired folded shape -> protein structure required to achieve this.
Based on what I've read, this reversed calculation hasn't been done yet. It might be one of those asymmetrical calculations that are much harder to reverse. (Giving rise to a new public key encryption method!)
The bigger issue is how available, costly, and user-friendly the hardware/ software is. Though unlikely to allow 2 – 4 guys in a garage to create medical/ materials sciences/ energy modelling breakthroughs weekly on their own custom rig – it does lead us to wonder how a company, university, or local enthusiast group could attain this type of computer power. Quantum computers are an obvious high-maintenance, specialist tech unable to get widespread availability soon – but a delorean-sized 100+petaflops 64-bit model with sub-military grade AI should be attainable in the early 2020s?
This is also one of the last critical technologies needed for molecular machines. Living things are built with self assembling 3D machine parts built out of chains of amino acids that are specified by chains of nucleic acids. The machines that do this work like Ribosomes and Cas9 are also proteins themselves built this way.
Being able to accurately predict folding of novel amino acid chains lets engineers build whatever 3D machine parts they can design as proteins with atomic precision.
This is a major step Eric Drexler was anticipating 30 years ago that’s just taken awhile.