2024 MIT Health Science Forum: Lightning Talk - qBraid

KANAV SETIA: So looking at the presentation, it seems like this may be a more computationally focused talk. So just quickly, how many people here have heard of quantum computers and no? OK. And how many of you all know about what they can do-- what algorithms they target? OK, very cool.

So just quickly, what are these machines? These machines are new kinds of computers that harness the power of quantum mechanics. This is the same quantum mechanics that governs the particles at microscopic scale. And these particles could be atoms and molecules, the systems you see in biology. And partly when you go and you try to simulate those systems using Schrodinger's equation, you require exponentially large computational resources.

So the same thing that makes it hard for you to simulate those quantum mechanical systems comes for free in quantum computers, which is why at a high level intuition-- this is the reason why you could expect exponential advantage in simulating these quantum systems using quantum computers, right? So now that you all are experts in quantum computing, let me tell you more about what we do.

So we have a one-stop platform that you can use to get access to all the quantum software and quantum hardware that's available in the market. So we work with over 50% of the quantum computing companies and make their software and hardware available in one location.

And you can access-- at this point, there's not a single architecture that is a winner for quantum computing. You have whole bunch of-- all sorts of quantum computers being made out of these superconducting circuits. You have atoms being used as qubits. You have trapped ions being used. So all of those are available for you to test your quantum algorithms.

And the way we work with the various different stakeholders is for the enterprise world, we provide them support for quantum algorithms. So if you are an end-user company, we'll work with you to come up with algorithms for your specific use case. So this is where we provide extra support. But overall, the platform is built for anyone who wants to use these machines and write their algorithms.

We also work with quantum computing companies to launch their software and hardware on your computer. So it's sort of like a marketplace-- a single place where you can get access to all the different quantum algorithms specific to-- and these are some of our clients and companies and partners who we work with. Specifically, I would like to talk to this audience about what quantum computers can do for the biotech world.

For the past year or so, we've been looking at this specific problem of beta amyloid that you heard today about. And the idea there is have this beta amyloid molecule, and it's been widely accepted for at least past 20 years. It interacts with metal ions, and it leads to formation of these plaques.

So the pathophysiology of this reaction is actually not very well understood as to where do these metal ions go and interact with what active sites on these beta amyloid molecules which lead to these plaque formation. So if we could understand this reaction, maybe there is-- might be an easy way to develop drugs for targeting that active site and thereby allowing-- not allowing the clumps to form.

And so again, you probably already heard, you already know, that over $100 billion have been spent on research understanding Alzheimer's and Parkinson's, with relatively low progress. And even this amyloid beta hypothesis has been brought under the question. And so our job is like, let's look at it from physics point of view to understand whether these protein molecules interact with certain metal ions and these lead to clumps.

So either answer would be very cool. And actually, NIH has spent $200 to $300 million on just computational studies. And it comes back to you do require exponentially large number of computational resources, and this is the problem that you can probably address much faster using quantum computers.

And the reason you have even machine learning models fail is because machine learning models need data, and data has to come from these quantum mechanical simulations. So if you can't even perform those simulations, where are you going to get your data? Or you need quantum sensors, which again, you may know they aren't ubiquitous or you don't have quantum sensors available in your body. So which is where machine learning models tend to fail as well.

So here's the pipeline we've been working on, and we have a full-functioning pipeline available here. What we do here is you have this protein molecule. And what you do is you fragment it into smaller pieces, and you recognize the piece that has some strongly interacting metal ions. And these are transition metals, which are known to be hard for classical algorithms.

So those are the algorithms. You send it to a quantum pipeline. The classical tractable ones you can simulate using classical molecules-- sorry, classical computers. And so again, what we do is we have a further fragmentation techniques, which are quantum in nature, that run on a quantum computers.

And so once you do that, you have good answers for the algorithms. And so quickly, the state of the field right now where quantum computers are at, the hardware needs to catch up. It's been making tremendous progress over the past five years, but it's still not there where it can run these large calculations.

So what we've come out with this innovative pipeline to use the current generation hardware and run these simulations. It's sort of like if I got maybe instead of 10,000 NVIDIA GPUs, what if I had one million GPUs? But you only have 10,000 to 100,000 GPUs, so your algorithm innovation comes in place.

It's the same idea. Our algorithm innovation enables you to use current generation of hardware and to get the best answers. And this is where we're at, and quickly we've got our whole bunch of data, which I'm not sure it's going to make-- oops, what happened?

There you go. So here, the major place where you should pay attention to is we reduced these numbers 27 10 raised to the 9th to 10. What that means is the circuit depth, the amount of calculation you had to do on a quantum computer, went from 10 raised to power, 9 to just 10. The number of qubits you required for a certain simulation went from 27 to just four.

And this is the innovation we've done over the past year that would enable you to target large molecules. And I'm happy to tell you much more about all the achievements we've had. And similarly, you can see the numbers here. You have 65,000 going to 2,500. So, again, I imagine because these may be foreign numbers to you, but the idea is we've had some really cool couple of orders of magnitude improvement in algorithm development where you can target many of the complex systems in the coming years.

And with these things. You can probably-- so when can you use quantum computers in your pipeline? Definitely before your next clinical trial finishes. So all right. That's my talk. Thank you.

[APPLAUSE]

Just a quick ask. We are actually looking for companies who would like to test this pipeline, and we're happy to work with you. Yeah.