When AI Meets DNA

DNA is hotter than ever.  We're doing more DNA sequencing to identify genetic risks.  We're using tools like CRISPR to "fix" DNA.  We've been using DNA to help identify criminals for some time, but now we're using relatives' DNA from ancestry sites to identify even more. 

Less than a couple years ago, using DNA as a storage medium was still at the laboratory level; now the first commercial DNA storage company -- a start-up named Catalog -- is set to launch in 2019.  Even U.S. spy agencies are trying to leap on the DNA storage bandwagon. 

If all that is hot, then here's what is really cool: using DNA as the basis for a neural network.  I.e.: AI DNA. 

Credit: The Sociable

Researchers from Caltech announced that they have developed a neural network made from synthetic DNA.  The network "learned" how to correctly identify handwritten numbers, a task that is not always easy for humans to do (as anyone who has to read my handwriting can attest).  The results were published in Nature. 

Lead researcher Lulu Qian explained what they did: "In this work, we have designed and created biochemical circuits that function like a small network of neurons to classify molecular information substantially more complex than previously possible."  

Translated, they created a molecular "smart soup" made of bio-engineered strands of DNA, and taught it to recognize handwritten numbers through a "winner-takes-all" process.  The neural network looks for certain concentrations of molecules and produces specified reactions when it finds them. 

Professor Qian elaborated to The Register:  

A single-stranded DNA molecule with just the right sequence of nucleotides can bind to another double-stranded DNA molecule that has a single-stranded tail. Once grabbed onto the tail, it can force the nucleotides in the double strands to open up, one nucleotide at a time, until the previously bound strand is released.

The invading strand can be seen as an input signal while the released strand an output signal, resulting in a simple input-output function. Once released, the output strand can then take on a different role as an input to interact with yet another double-stranded DNA molecule, leading to a network of molecular interactions that compute more complex input-output functions.

It looks something like this:

Source: Nature

OK, it's not pretty, but it's pretty impressive.  

Qian and first author Kevin Cherry, a graduate student of Professor Qian, plan to expand their work to have the neural network to form "memories" from examples added to the test tube, allowing it to be trained to do a wider range of tasks.  As Mr. Cherry sees it:

Common medical diagnostics detect the presence of a few biomolecules, for example cholesterol or blood glucose.  Using more sophisticated biomolecular circuits like ours, diagnostic testing could one day include hundreds of biomolecules, with the analysis and response conducted directly in the molecular environment.

Right now, the DNA neural networks have a limited set of tasks they can accomplish, and the computation using chemical processes is much slower than "traditional" computing.   Still, Professor Qian sees the potential:

Similar to how electronic computers and smart phones have made humans more capable than a hundred years ago, artificial molecular machines could make all things made of molecules, perhaps including even paint and bandages, more capable and more responsive to the environment in the hundred years to come.

People have been talking about nanobots in healthcare for many years now, with several interesting applications already being tested and some typically optimistic predictions about the market potential, but we may have been thinking about them all wrong.  Instead of tiny versions of traditional computers, perhaps built with organic materials, they could be DNA-based neural networks.  The possibilities truly are mind-boggling.

If you think all that is far-fetched, U.S. spy agency effort -- Molecular Information Storage (MIST) -- referenced above calls not just for DNA-based storage and retrieval, but also an operating system. 

The thing to keep in mind is that, although the processes DNA uses may be slower (now) than traditional computing, the storage capabilities are exponentially greater than the methods we use now.  Hyunjun Park, CEO and co-founder of Catalog, told Digital Trends: "If you're comparing apples to apples, the bits you can store in the same volume comes out at something like 1 million times the informational density of a solid-state drive." 

Credit: Pasieka/Science Photo Library, Cosmos

As I put it before, you could literally be your own medical record, using DNA storage.  Mr. Park seems to agree, noting:

Imagine a subcutaneous pellet containing all your health data, all your MRI scans, your blood tests, your X-rays from your dentist...If you had that with you in the form of DNA, you could physically control that data and access to it, while making sure that only the authorized doctors could have access to it.

With the new work from Caltech, now I'm wondering if we could be our own EHR as well -- not just the data but also acting upon it.  A DNA-based computational device using DNA as the storage medium, stored within us -- possibly even encoded within our own DNA. 

Mind.  Blown.

Here's an even more out-there idea: maybe we could "teach" our microbiome to speak up for itself and tell us how we can help it help us better.  Think of how that could improve our health.

We are, in many ways, still in the first generation of computing; conceptually, modern computers are not really different than the bulky computers of the 1950's -- just much, much smaller and faster.  The next generation may use approaches like quantum computing or distributed computing -- or perhaps DNA computing.

Similarly, we're barely in the first generation of artificial intelligence, but we've been building it using our traditional concepts of computing.  That is certainly going to continue to evolve, rapidly, but we should also be thinking about how and when DNA-based AI might be more applicable, especially for healthcare. 

We're a long way from a robust DNA neural network, much less a true DNA AI, and who knows where they may lead, but I, for one, am going to be watching closely.