 Thank you. So good afternoon, everyone. The main focus of my talk is going to be on how we can take a more data-driven approach to food safety. And by food safety, what I really mean is bacteria. So the harmful bugs that get into the food chain and cause foodborne illnesses. So it's kind of unfortunate timing that my talk is so close to lunchtime. I'll do my best to not put everyone off their food. But to go straight into the scaremongering and put a bit of context, even already this year there's been a lot of high-profile incidences of bacteria being caught in various foods. So one in the US with Listeria and Salmonella in ice cream. A really devastating one with infant milk formula where Salmonella was found and it had a huge impact mainly in France and Belgium at the start of the year. Or even a bit closer to home where there was a recall for mustard seeds with Salmonella in them. And again, I mean, these products, ice cream, seeds, we think of Salmonella, we think of chicken. Everyone knows you need to cook your chicken well. And you kind of take it for granted that everything else should be clean. But obviously there's a bit of work that needs to be done still to make sure that bugs aren't still getting into the food. And the way it's happening at the moment still is that a lot of the techniques that food companies are using to make sure that there's no harmful bacteria, it's kind of the same techniques that have been around for decades. So they take a sample and try and culture or grow a particular bacteria in a little Petri dish. But it's very much a one-test per-bacteria approach. And as we understand that more and more different bacteria become implicated in illnesses, that stack of Petri dishes needs to grow and grow and grow. It doesn't scale very well. In addition, there are so many bacteria that we actually can't even grow in this way. So the test for them is really hard. But thankfully we can use genomic sequencing techniques to kind of bridge this gap. And there's a whole range of flavors of genomic sequencing. But the one we're going to focus on is using an approach to basically scoop up as much DNA from a sample as we can and target these little fragments that are enough for us to understand what bacteria were in that sample. So we want to build up a picture of the profile of bacteria in some environment, or what we call the microbiome. And really the microbiome is just all of the little microscopic bugs that live in some environment. We'll just focus on bacteria for simplicity today. But I mean, everything from the stage I'm on, my skin, my stomach, they all have their own particular characteristics, their own little family of bacteria living in them, their own microbiome. And it's a huge amount of research booming at the moment around understanding the microbiome, particularly for humans and implicating the bacteria in your gut with various diseases. But there is a lot of work in terms of doing more environmental monitoring and looking at how bacteria in facilities can have implications for whatever is going on in that region. It's even made it all the way up into space. So the ISS, there was a paper released earlier in the year giving a breakdown of the bacteria that live up there. Apparently, it's absolutely filthy. But if we can do it in space, we can definitely do it in a food manufacturing plant. And in terms of how we get that data, it's quite a significant pipeline to actually extract some data from a sample at the end. I won't go into all the steps, but just to point out that the actual sequencing bit, there's actually a kind of a threshold in how much DNA we can get back. So we always really think of it as being like a random sampling from the DNA that's in the population. And we build up then a picture of which bacteria are there, and we normalize it so we get something like this. So this is just one particular sample, and we can see that it's made up of 10% of lacticoccus, for example, here. So lacticoccus, you'd find it in milk. So this is probably something that came from dairy. And we can, of course, group the bacteria by families and so on, which is what those concentric rings are. It's kind of like a hierarchy going deeper and deeper into the center. Now, we've kind of pulled together from a number of studies, you know, sort of to get a rough ballpark. On average, we see about 200 different types of bacteria per sample. So that's just from one tiny little swab of some region, a huge number of different bacteria. So there's a lot of information here, but I mean, remember, this is just one data point, essentially. So I mean, sure, it's good knowing that bacteria A, B, and C are in my sample, but really, it only becomes useful when we actually start to gather lots and lots of samples from different sites and locations to build up a big picture of what's happening inside our plant. So here's a really low-fi, rough schematic of an actual food producing facility. And each of these points are locations where we've taken samples from the environment and studied them. And the color represents the intensity or the abundance of one particular bacteria. And this bacteria actually grows mostly on your skin. And it's the reason why in any kind of sensitive area, people have to wear gloves, gowns, hats, and so on to stop the spread of this. And that yellow box is the changing room where the employees go in to suit up before they actually go and handle the food. So we've got a nice validation of the fact that this is actually working to reduce the spread of this bacteria from skin. And you see that the intensity goes down as we get further away from the changing rooms. You can also look at bacteria over time, of course, a number of different locations. This is a different bacteria. It likes warm conditions to grow. We've got a couple of locations here where there's a little spike at the end where this guy suddenly started to thrive. And that actually correlates with the glorious summer that we actually had for once last year. It seems like a lifetime ago now with the weather outside. But so those three locations where this bacteria is rising in abundance, clearly they're locations where the controls to prevent the outside environment, so the warm weather affecting the interior of the plant, are lapsed. And this issue of control is really important, because these two examples even that I've just shown are still looking at bacteria in isolation. But really, we want to be looking at the entire microbiome because for a controlled plant, it should be really stable. So here we're looking at one particular location over 14 months, regular sampling, and you can see that it's pretty stable. So each of these colors is a different bacteria. The teal bit at the top is a grouping of all the really, really low abundance guys that only appear in tiny amounts. They're much more transient, so we do see fluctuation there. But for the more dominant guys, they're pretty much set. They live there, it's their home to stay the same. So rather than looking at individual bacteria, we can look at when this environment is changing. And we can do that using some anomaly detection. So this is where we bring machine learning into it, and we look at when the microbiome is starting to deviate from what we would expect to see in that location. So this chart here is just topological mapper applied on top of some anomaly detection, and we can see that there's two clusters here that have been separated on the right. So there are groups of samples which look different than what we would normally expect, and warrant some investigation. So clearly the environment has changed to allow the microbiome to deviate. And once we do see deviations, because we have so much data, it becomes much easier to actually start to track and understand where contaminations have come from. So here we again, we have a number of samples coloured by date. So again, this is a regular periodic sampling. And those red guys kind of stand out on their own. So we had some period where something happened and the environment changed. If we shift our focus and instead look at the type. So we have samples, some of them the blue guys were swabs from the walls, the ceilings and so on. And the multi-coloured ones at the bottom were various product samples. So food at various stages of processing. And the one that really stands out is this little red guy here because he came from raw input that was brought into the plant. So this is an ingredient that was used to make the final product. So it hadn't had a chance to absorb the microbiome of the plant. Rather, it had the opposite effect. This is the source of the contamination. And again, because of contamination, it's not going to be one or two bacteria. It's going to be an entire community of bacteria that are brought in. And we can see how this then propagated among the plant. They just want final point while I'm talking about comparing different samples. Again, this is just a simple PCA of a number of different processing facilities. So different food producers, each one's a different colour. And we can see that they all kind of cluster separately. So in other words, the microbiome in each of these locations is different. And it's something that we've seen quite a lot in, again, the human microbiome research where people now can understand that everyone's got a unique gut microbiome. And there's even companies offering tailored dietary recommendations based on taking a sample and tailoring it to the unique microbiome that lives inside you. But the same principle should apply for any kind of microbiome or community of bacteria. So for these food producers, each of them have different bugs living in their plant. So when it comes to controlling or designing a sort of a safety procedure, there's no real sort of one-size-fits-all, right? And again, so we can take a kind of personalised approach to designing a safety control by understanding what the typical background bacteria are for that particular plant. So a lot of this is very much sort of retrospective kind of analysis. You know, we've got a lot of samples. We've analysed them. That's all nice. But I mean, going back to those anomalies, if we are contaminations that we might identify, suppose that's a big colony of salmonella that we find inside a food that's just about to be shipped. So of course, that's in the bin. It's gone. We've made a loss on the processing for that. We really want to try and identify upfront when we're going to see bacteria. And there's kind of some latent information in all of this data that we've gathered that we haven't really tapped into. And it's the fact that we're looking at communities of bacteria. So we can start to piece this together and understand the relationships between those bacteria from all this data that we've gathered. So this is just a very, very tiny snapshot of a much larger bacteria network. So each of these nodes is a distinct type of bacteria. And they're joined here based on how likely they are to appear together, based on all the samples that we've gathered through monitoring programs. I'm just focusing on this little bit because that yellow guy there that I've circled is Listeria. So Listeria is one of the deadliest ones that food companies are always on the lookout for that should not make it into a final product. And so we can see the Listeria is kind of on the edge of this really tightly connected cluster of a handful of other bacteria. And it's so dense and tightly connected because these guys really tend to hang out together. So if you see one or two of them, chances are you're going to see all of them. And this allows us to get some early indicators of when we're going to see Listeria because we now have other bacteria that we know tend to appear when Listeria does. And so by combining all these relationships with a predictive modeling framework, we can build models which will predict the likelihood of seeing, in this case, Listeria based on the existing bacteria that are there. And then we can track that over time and manage that risk allowing us to take preventative action hopefully before we actually do see Listeria. Okay, so just to summarize all of that, I mean this kind of thing, the microbiome research is booming at the moment because genomic sequencing is becoming really cheap. So this kind of monitoring approach is actually becoming much more accessible. And of course once we have that data, I mean you've seen really simple analytics can start to give us a nice picture of what's happening in our environment. But the real cherry on the cake then is a fair relationship about the bacteria themselves and which bacteria are, you know, they like to hang out together or which are antagonistic, which allows us to bring predictive models into play and be more preventative in managing risk. Thank you. Thank you, Noel, for that. I suppose one quick question. I think it's fairly clear from your talk I suppose the benefits that are there for food manufacturers. And I suppose obviously they're keenly aware of the risks involved and ultimately the financial risks involved in having let's say a listeria outbreak. But I imagine there are still some challenges in this being an early case study. So could you maybe tell us about the barriers to bringing this through to as opposed to every food manufacturing plant in the world really? Yeah, well I guess probably the biggest one is regulation at the moment. I mean obviously there's laws around, you know, tests that have to be done to ensure that there aren't bacteria getting to the food. But they're still the old-school approach of, you know, tests for bacteria, A, B, C, D, and so on and give us the results. So because the regulations aren't really changing, you know, companies are kind of reluctant to change their own approach. But I mean this kind of monitoring and genomic sequencing to understand the microbiome is pretty common now. And even actually the FDA and the US are doing huge amounts of sampling and looking at the bacteria in various environments. So of course the FDA, being the regulator in the US, chances are that maybe this is going to become something down the line so companies should be getting more involved. I think the one thing that's missing is that even for the monitoring that's happening at the moment, it's still very biology-focused. You know, they get a sample and they see what bugs are there, and that's kind of it. There's kind of the missing piece of bringing, you know, data analytics on top of it to actually start to get some real insights. I think that if we can get that piece there as well, companies will see the benefit and start to adopt more. Sure, I suppose get ahead of the issue before it becomes an issue really. Very interesting. No rogers, everyone. Thank you very much.