 Good morning, ladies and gentlemen. It's a pleasure to attempt in 14 minutes and 50 seconds to describe some of the potential of predictive analytics in the area of food science, especially in the area of food safety. So I won't waste any time. But in walking into the RDS this morning, I was reflecting on the fact that the first time, the last time I was here, was as a 16-year-old for the Young Scientist exhibition. And it was at the time sponsored by Erlingus. And I had a data problem. I won't tell you when it was, because it was a very long time ago, but I had a data problem. It was a project, an observational project on water pollution. And I had to crunch some data. I didn't want to do it by hand, there were no calculators. I had to go to Cork to rent a calculator that was capable of calculating standard deviations. That's how far we've come since Erlingus sponsored the Erlingus Young Scientist competition a long time ago. So with that, I'll move on to give you an outline of what I want to talk about. So I started my professional career, my postdoctoral career in predictive toxicology. And I worked in computational chemistry as a postdoc. And so one of my favorite topics is predictive talks. And I want to use that as a case study this morning. And then the second area is in relation to text data, which is in early warning, food safety early warning systems in the food industry. And I'll give you a brief outline of that. The changing face of lab data, and this will be your entree to Professor Fanning's talk, who's going to talk more about that. And then life in the dry laboratory, because nowadays, many of us exist in the dry lab. We don't connect with the wet lab anymore. And that's where some of the most exciting stuff is done. And then some caveats, challenges, and solutions. So we're living really in an area of unprecedented change. We have more data than we know what to do with. We can't process all the stuff we have. So the models are taking time to catch up with the availability of data. And so that's where most of the work will have to be done. And in parallel with that, we've got formidable, double challenges, especially in the area of food safety. We've had FRIP for a little in the last few weeks. We've had the chicken problem now in the UK again. So every week brings a new problem. So the problems are not going away, despite the fact that it's often said we're at peak certification right now. We've got BRC and ISO certificates and any other kind of certificate you want. And yet, the problems are still with us. So let's talk a little bit about what we're doing in my company and other companies in the area of predictive toxicology. It's very basically starting with chemical structures, classical two-dimensional chemical structures, and transforming those structural data into different types of representation of the molecular properties. That could be volume and surface potentials. It could be hydrophobic and polar regions of the molecules. It could be hydrogen bond potentials, hydrogen bond acceptor potentials, and then electrostatic molecular potentials as well at the end. This last one, of course, being based on molecular orbital calculations, gives you a more of a picture of the electronic environment that the molecules are generating and experiencing. Food safety early warning. Well, as I said to you, we've got a lot of issues coming down the line. And what we have to do is try and look further ahead. We've got to try and be on the front foot and anticipate what's coming next. So how do we go about that? Well, we use Patrick Lagodak's approach, which is very much looking at the different types of evidence that if you don't process or don't attend to, turn into incidents and ultimately into a crisis. So if you look at the early signals, you process the early signals, you can sometimes prevent these crises happening in the first place. So that's the whole principle of it. The way we do it in the Nestle organization is to start with a human network, 150 people. Much has been said this morning about the value of judgment. So judgment is still important. The data are important, but you need humans with the data to optimize what you do with it. So you need actionable data. And to do that, we use an early warning expert network of 150 people. It's a multidisciplinary team. It exists in R&D and operations. And that's complemented then by web scouting, which looks at about 8,000 websites a day. About 300,000 articles are scouted per year. 10,000 articles are kept on a database for future reference. And then we disseminate the information throughout the organization. Much has been said this morning in the panel discussion about social media. Yes, many companies are monitoring the social media. We can track what's going on. But again, we're more or less looking backwards. We're not yet very good at actually using it for future decision making. So it's all really about continuous change needs continuous thinking. And then you need goal alignment. You need aligned organization so that common team thinking is needed throughout the organization to keep things together. Just to give one example of where this type of thinking is employed, we had a problem in Malaysia where we wanted to support farmers who wanted to increase their yield of rice from about 2,000 kilograms per hectare to 8,000 kilograms per hectare. And what we actually started to see, of course, was that we saw arsenic levels in the rice getting higher. This was a trend analysis of existing data. And we can do that because every company generates a lot of analytical data. In the case of Nestle, it's 130 million data points per year. And the data are volatile. We use them, we make a decision, and then they're left there in a database. So the question then is, can we use the data? Can we mine the data more effectively to prevent future problems? And this was what happened with arsenic. We saw the trend, it wasn't in the danger zone, but if it continued, it would have been. We would have ended up being either out of compliance or in an unsafe situation. So working with the farmers, we actually identified the fertilizers to be the main cause of the problem. And working with the farmers, what we were able to do was to allow the increase in yield without actually causing a safety problem. This was a classical win-win situation, optimized yield for the farmers, optimized safety for the manufacturer. Steve Jobs said that we cannot join the dots looking forward. We have to always join the dots looking backward. Bill Gates made a similar observation. So when we are actually talking about early warning and looking to the future, we actually have to look in the rear view mirror a lot of the time. They say that foresight is hindsight plus insight. We've got also to address the history and analyze the history before we can actually make good predictions about the future. So effectively, this is what we're doing much of the time. What about the lab data? You've all heard about Moore's Law. How many of you have heard about Carlson's Curve? Carlson's Curve effectively is the equivalent in the biological sciences, and it refers to the decreasing cost of genome sequencing. Here's the curve. It's decreasing very, very dramatically from the early point when in around 2001, it was about 100 million to sequence a genome. 2008, I think it was still around 10 million to sequence a genome, and today it's about $1,000 to sequence a genome. So again, it brings about huge, huge access to data, and Professor Fanning will actually go into more on that. Also, of course, the equipment is getting cheaper. We've got min-ion. We've got the high-sex and my-sex, which are becoming routine lab equipment, but we've also got personalized equipment that you can take into the field, and of course the Ebola crisis used these in West Africa to do a lot of the field work. So you can get the data very easily, even if the data are difficult to analyze sometimes. Then we've got genomics tools like the ion torrent from thermal systems. These are nice because you can go on a single chip. You can actually analyze 340 samples at a time. There's about 3 million wells on the chip. Each of them is effectively a small pH meter, measuring a release of a hydrogen ion. So again, you've got a phenomenal potential here to generate data. So we're moving into dry lab, and what do you do in the dry lab? Well, people have talked already this morning about literature-based discovery. There's examples where IBM's Watson was used to mine. I think it was 300,000 oncology papers and actually made a new discovery based on rereading those informatically. So there is potential for literature-based discovery. If you want to look at the actual potential, you can look at the Avicen, a roadmap online, and for them, they're looking at in-silico clinical trials that can help to reduce, refine, and partially replace the human trial. You don't even have to do the human trial. You can use a model instead. There's a lot of interesting discussion right now about post-diction analysis, where you add data and you re-crunch the data after the actual trial is done. So you can do a lot with all data. You don't have to throw it out once you've actually completed the study. Prediction is a word I have great difficulty with. I've said this to Kian already. If you want to be on-kind, prediction is fiction. In fact, what we're often talking about is extrapolation, and probably have to be more precise about what we're actually doing here. Are we predicting or extrapolating or interpolating? But in-silico, we can actually do a lot of help. We can clarify. We cannot express it with a number. We don't know much about it, as Lord Kelvin said. So we can actually crunch more numbers. We can actually optimize the experimental design before we commit to the expense of a clinical trial or a lab study. We can actually test hypothesis in-silico. We can test the sensitivity of our models. Biology is complex, and I think increasingly with food and biological sciences, we don't have to actually manage all of the complexity. We can actually take out some of the complex elements and variables and actually do the use models instead. I can give us qualitative insight. In-silico is cheap. It's fast. It can be used, of course, as well with experts to elicit the judgments of the experts bring. We're doing a lot of safety evaluation now based on in-silico. As I mentioned with predictive toxicology, the same is true with predictive micro, where you model outcomes. And we're getting to the notion of the virtual asset increasingly. If you've got enough data, you can actually model your biology. Now, moving to the caveats. As I said, I would discuss caveats and solutions. Well, unfortunately, it's been said already, data are not information. It's not the same thing. So having more data doesn't mean we know more. And as Einstein said, information is not knowledge. And more importantly, knowledge is not action. So actionability is important. We cannot live in a world where we simply know things. We must do things. That's what's important. We're not doing something with what we know and what we have, then we're not actually being successful. And I think increasingly as we have more data and more knowledge, we need to organize it better. So the models and the structures are going to be of paramount importance. Edmund Spencer was a British philosopher, lived around the turn of the 20th century, beginning of the 20th century. His famous observation was that if a man's knowledge is not an order, the more of it he has, the greater will be his confusion. That's where we are sometimes. We're confused by having too much information. And the decision-making ultimately will be based on knowledge, not on numbers. So what we actually have to encourage now are data translators. We need a narrative on our data. Some people talk about let the data tell a story. The data can tell a story. The story is told by the people who actually interpret the data. So the data narrative is going to become much, much more important as we generate more data. And many of you have read The Black Swan, I guess, and Fooled by Randomness, both of whom were written by Nikola Taleb. I think it was in The Black Swan that he alludes to the illusion of understanding. You know, sometimes we fool ourselves in thinking that we know a lot more than we do. Richard Feynman famously observed that the first rule of science is not to be fooled by not to fool yourself. The second rule is to understand that you're the easiest person to fool. So sometimes we fool ourselves with our own data. And so the solutions then, coming to the end of my brief presentation, the question we have to ask is, are we really predicting or are we extrapolating? Cronin actually said this morning, he mentioned the issue of traceability, transparency and validation. We need to ask those questions. Is the process traceable? Is it validated and transparent? We need to be aware of the realities of p-hacking and regression to the mean. Anybody had a p-hacking here? Some of you have. Well p-hacking is when you have a lot of data and you look for the p-value that's significant and you'll find them. If you're crunching off, you'll find p-values that are significant. Well so what? People call it p-hacking. It may be intentional, it may be accidental, but it's a problem. And then what's the framework for interpretation? Again, you know, the structure, the expertise structure. And is there a data narrative and do we understand the underlying biological systems? And to share with you my final slide, much of the opportunities I think in terms of solution finding will be around systems analysis and network science. These are two examples where the complexity has been reduced into visual examples. One is the obesity map produced by the Department of Trade and Industry in the UK. And then on the right hand side was a paper written by Joseph Barani and his colleagues at the Institute of Food Research in the UK where they're looking at mapping food trade systems. And again, it helps to visualize data so that decision makers can actually make sense of it. So that concludes my presentation. Thank you very much for listening.