The Story Behind Serox: How Our Mission Began

[00:00:00] Susannah de Jager: Thank you for joining this miniseries of Oxford+ with me, Susannah de Jager and Cici Muldoon. Over the course of the next eight episodes of the miniseries, we're going to discuss all elements of application, mission statement, investors, partnership, and place that make up a successful company, and we really hope you enjoy it.

Cici, you and I have spoken about your company's opportunity to disrupt point of care diagnostics. But let's take a step back and just hear a bit more about the underlying technology that you're applying.

[00:00:37] Cici Muldoon: Sure, so what we are doing is we are using advanced photonics, and machine learning, to analyse urine and classify it as diseased or not diseased across a wide variety of indications, including more than 20 different types of cancers, and this is meant to be a triage tool that gives you a red, amber, green as to whether you need to go down a further diagnostic pathway or you do not.

[00:01:00] Susannah de Jager: And when you talk about photonics, can you give a little more color on the specific area that you are using?

[00:01:05] Cici Muldoon: Sure, so we make use of something called spectroscopy. Spectroscopy is essentially the study of matter with light. So it's using light to analyse different substances. There's many different types of spectroscopy out there. We use a particular type of spectroscopy called Raman spectroscopy, which looks at the so-called vibrational and rotational modes of motion of molecules, and within that we focus on a particular type of Raman spectroscopy called Surface-enhanced Raman spectroscopy, where you use metal nanoparticles to essentially amplify and intensify the signal that you would get from all these molecules, and you can essentially tailor the signal so that you get a little bit more emphasis from some molecules in others.

To contextualise it, what we do is we take a molecular soup, which could be urine, it could be plasma, it could be saliva. We shine laser light at it, and when the light shines impinges on all these different molecules, all the different molecules will interact with the light in a different way and so the signal that you gather back will have a particular signature that comes from every molecule, and with the SERS we're able to intensify some of the signals more than others.

[00:02:18] Susannah de Jager: Can you explain a little bit more how you're actually using that to identify diseased samples?

[00:02:25] Cici Muldoon: In simple terms, your body is constantly shedding bits and pieces, so all your cells are shedding bits and pieces of DNA and so...

[00:02:32] Susannah de Jager: Such a nice thought.

[00:02:34] Cici Muldoon: And all those things are coming out of your body in your waist disposal. So all those things are coming out of your body, either in your urine, your saliva, your sweat, your breath, and your poo.

Now, if I were to look at your urine today, essentially I'm getting a snapshot of the waist disposal of your body. All the different things that are coming out will have a particular level. Now when you get sick in a particular way, the level of the different things coming out in the waist disposal changes. Now, depending on where you get sick, the levels change in a different way.

What we look at is the aggregated response of all the different bits of the waste disposal in your urine, of all the molecules in that molecular soup, all the biomarkers, and we compare what the pattern looks like when you're healthy and what the pattern looks like when you're not.

So what we essentially do is we take a urine sample, we put it onto a cartridge, onto which our nanoparticle substrates have been printed. Think of it like an LFT test. We put it into a spectrometer. Light shines onto your urine, all the different molecules in the urine will interact with that light in a different way. They will be amplified selectively by the SERS substrate and then be captured back.

Now, once we have these patterns. We break them down mathematically, using machine learning, and we plot them in a dimensional space. So imagine I test 350 samples of diseased urine from males that have prostate cancer around 40 years old, and then I test 350 samples of 40-year-old males without prostate cancer for argument's sake, and I get a pattern from each one of those. So I get 350 disease patterns, 350 non-disease patterns.

I break each one down mathematically, and I end up with two clumps, 350 disease, 350 non-disease. So I now train my machine learning model, and I say, I can give all the metadata I want, but I train this machine learning model and I say, this is what prostate cancer looks like. This is what not prostate cancer looks like.

I'm simplifying it hugely, but this is the way it works, and now if you come in with a mystery sample and you say, does this person have prostate cancer or not? If it falls in this lump they do. If it falls in that one, they don't.

[00:04:54] Susannah de Jager: So it's like a molecular fingerprint that you are creating?

[00:04:57] Cici Muldoon: Yes. It's essentially, it's molecular fingerprint.

[00:05:00] Susannah de Jager: But you didn't first use this technology in this application, so can you tell me a little bit about how the idea evolved, and I think it's quite fun to hear where it started.

[00:05:09] Cici Muldoon: Sure, absolutely. So I mean, as I mentioned, urine is a molecular soup, as is plasma, as is saliva. It's what I like to call a highly complex dilute liquid, and there are many highly complex, dilute liquids out there.

The first one that I focused on was actually wine. So my first business VeriVin analysed, unopened bottles of wine through the glass to determine whether the contents were, what it said on the label or not. And what we were doing was, essentially analysing a molecular soup, except that in this case it was mainly water, followed by ethanol and then a thousand different organic molecules that make up a wine.

The challenge then was being able to do this without taking the cork out, looking through the glass, but the methodology was much the same. It was using Raman spectroscopy in order to get a molecular fingerprint of the contents of the wine. Looking at all the ensemble of molecules together, using machine learning to categorise that signal mathematically and then classify it, in this case as 1985 Lafitte or not, or as all the same wine, or not the same wine. So it was still using Ramen spectroscopy and machine learning just in a very different context and a very different molecular soup.

[00:06:27] Susannah de Jager: I love that. So it's obviously got this exceptionally wide possibility of application. What made you pivot?

[00:06:35] Cici Muldoon: You've hit on something very important, which is that the technology is biofluid agnostic and disease agnostic, and if you even go one step further, it's sample agnostic and it's classification agnostic. So in one case we were doing wine and is it counterfeit or not? We are doing urine, is it disease or not?

The journey there was an interesting one. So we built a fantastic technology platform focused on wine. Unfortunately realized that there was, not really the market need that we expected. We along the journey had tested a lot of other different liquids, including Manuka honey, olive oil, injectables, perfume, and we could find business cases in all of these and some market need, but none of them were really a big enough business case or compelling enough.

Eventually, the idea was post-test to look at blood transfusion bags to see if they were past expire or not. So using the same type of, non-invasive spectroscopic analysis to look inside a blood bag and see whether it was still viable or not. So this led me down a path of doing some research into what was being done using Raman spectroscopy and machine learning looking at blood. Which eventually landed me in a bunch of literature where people were using Raman spectroscopy and a variant thereof called Surface-enhanced Raman spectroscopy to look at urine and again, analyse it for the presence of different disease states, and this was for me, an absolute no brainer to pivot the company towards something that could have such a huge impact and that had such a compelling business case.

[00:08:11] Susannah de Jager: Cici, thank you so much. That's a really wonderful explanation of the technology that you're applying to this extraordinarily important question of how to identify disease in the easiest way possible.