Exscientia Plc

Hello, everyone. My name is Chris, and I'll be your conference operator today. At this time, I'd like to welcome everyone to Accenture's business update call for the fourth quarter and full year end of 2021. All lines have been placed on mute to prevent any background noise. After the speaker's remarks, there will be a question and answer session. If you'd like to ask a question during this time, simply press star then the number one on your telephone keypad. To withdraw your question, please press star one again. At this time, I'd like to introduce Sarah Sherman, Vice President, Investor Relations. Sarah, you may begin.

Thank you, Operator. A press release in Form 20F was issued yesterday after U.S. market closed with our fourth quarter and full year 2021 financial results and business updates. These documents can be found on our website at www.investors.accentures.ai, along with a presentation for today's webcast. Before we begin, I'd like to remind you on slide two that we may make forward-looking statements on our call. These may include statements about our projected growth, revenue, business models, and business performance, including with respect to our technology platform and pandemic preparedness program. Actual results may differ materially from those indicated by these statements. Unless required by law, Accenture does not undertake any obligation to update these statements regarding the future or to confirm these statements in relation to actual results. On today's call, I'm joined by Andrew Hopkins, Chief Executive Officer, and Gary Pardo, Chief Technology Officer. Ben Taylor, CFO and Chief Strategy Officer, and Gabe Hallett, Chief Operations Officer, will also be available for the Q&A session. And with that, I will now turn the call over to Andrew.

Thank you, Sarah. And thank you to everyone who joined us today. 2021 was a remarkable year for Accenture. We strategically scaled the company and we expanded our capabilities. As you can see on slide three, we have significantly grown our pipeline year over year, adding 11 programs and advancing two programs into late discovery and three into IND enabling studies. The press release issued last night included an exhaustive review of our 2021 accomplishments. Let me recap a few of the most notable and a few recent highlights. Hiring key talent and expertise, tripling the size of our global workforce, and adding to our U.S. footprint with a new Boston office and office expansion in Miami. Completing our acquisition of AllSite, integrating the world's leading patient tissue platform into our end-to-end system, and gaining a tremendously talented team. listed on NASDAQ and raised in over $510 million in gross proceeds from our IPO and private placement. We ended 2021 with approximately $759 million in cash or cash equivalents, and we are well positioned to deliver on our strategic imperatives. Announcing one of the industry's largest AI-powered drug discovery and development deals to date. We've got a $5.2 billion collaboration with Sanofi with a $100 million upfront payment. Successfully executing our partnerships, as we've seen by the expansion in work with three of our major partners, BMS, Sanofi, and with Bill and Melinda Gates Foundation. With BMS in licensing and AI-designed immunomodulated drug candidates. The successful application of artificial intelligence and machine learning to reduce our industry's failure rate and produce better, more effective medicines has long been recognized as transformative potential. We are now working to put that promise into practice. In the last several months, we've seen some of the world's largest drug makers announce the largest deal to date in AI-powered drug discovery. The back-to-back announcements by Titans and Biotech and Pharma represent the industry's fullest embrace of AI to date. And we think of an inflection point in the evolution of AI-powered drug discovery in Iran. It should come as no surprise that there's mountains of satisfaction with the time it takes to deliver new medicine, particularly when we are faced with urgent health crises such as the global pandemic. Even more frustrating is that most of this time is spent trying to fix problems as they arise, It was currently a lengthy, step-by-step process, drawn out over the course of 10 years. It's striking when you consider that no other consumer products are made by this way. By the time a drug reaches the patient, the underlying research is dated by 10 years, and the science is likely significantly advanced. Can you imagine if any other technology products were made in this way? The founding team and I set out to build a completely new type of company, to re-engineer the drug discovery and design process. Today, that's best illustrated in the near equal split in our team between drug discovery scientists and technologists, which you might be surprised is an anomaly in our industry. By bringing together these two seemingly desperate disciplines, our scientists are able to tackle new problems with the power of our AI systems, while our technologists encode these learnings, working towards a day when we can achieve full automation. Today, Our chief technology officer, Gary Paradue, will talk more about our technology and how we're using it to design and develop better molecules and the deep investments we've made in technology. What I find incredible about this is how our underlying technology and AI platforms may have the potential to achieve feats that were never seen before in drug discovery. Our AI platforms can make decisions based on analyzing thousands of different parameters in parallel, enhancing creativity with generative algorithms, working in a computational space far beyond the ability of any one scientist or team of scientists to consider. Our drug design process, from the AI generation of the first novel molecules to the design of a development candidate, has averaged about one year versus industry standards of four and a half. Our AI-driven methods lead to the nomination of drug candidates after the synthesis, on average, of less than a tenth of the number of compounds versus the industry average. This efficiency enables us to concurrently advance more than 30 programs. I'd like to think of AI as supercharging our amazingly talented drug discovery teams. This is a combination of human and machine that's enabled us to begin to crack forms and areas such as truly personalized medicine, an area that our industry has been talking about for more than 25 years. as seen by results published in Cancer Discovery, where a platform was the first to successfully guide treatment outcomes for late-stage cancer patients, achieving a 55% ORR. This gives us confidence that the models we're developing may translate to potential patient benefit in the clinic. This is an area that I'm personally very enthusiastic about, and I look forward to seeing where we can take the platform next, including ovarian, lung, and breast cancers. And as we look at what's ahead in 2022, we're driven by the possibility of how much we can advance, powered by this AI-led approach. We anticipate continued expansion of our pipeline by not only adding new discovery programs, but by also continuing to nominate new drug development candidates and progress them towards the clinic. So the ability to have clinical capabilities and infrastructure Increasing validation of a platform for additional data, including data on our pipeline programs EXS21456 and GTAEX617 that will be presented in April at the upcoming AACR Congress and throughout 2022. Further, our mission is to fully automate drug creation for the opening of our laboratory automation suite in Oxford. Today, Gary, our CTO, will be focusing on just one aspect of our tech. How do we design better drugs? The technology team is up to some incredible work this year, including opening and operationalizing a new 26,000 square foot automation suite that will bring us one step closer to what's fully automated in the chemical synthesis and analysis of our small molecules and our drug discovery programs. And I'll now turn over the call to Gary to walk through our technology platform. Thank you, Andrew. Today, I would like to give you a high-level overview of our technology platform so that we can bring to light how the underlying technology at Excientia is differentiated from what others in the industry are doing. There are several fundamental ways in which I believe we stand apart, but perhaps the easiest way to explain it is where we start, with the patient, as you can see on slide six. We think about drug discovery as a learning cycle. A cycle that begins with the patient, fueled by our AI platforms, that enables us to learn from every new piece of data and bring more information to bear through every step of drug creation. In a conventional drug discovery project, it may take years before a potential new drug candidate is tested in humans. With our AI precision medicine platform, we're able to bring this process much, much earlier into the discovery phase. On the next slide, we show how we are identifying the bright target. This is possibly the most important decision for a drug discovery program. We use Stentor Biologist, which integrates literature along with genomic and transcriptomic data into our knowledge graph to identify connections and predict target to disease associations. This process is disease area agnostic, with application today across oncology, immunology, immuno-oncology, and rare diseases. Our precision medicine platform utilizes primary human tissue samples, and we align our early target identification activities to leverage this platform, capturing the insights from drug action on patient cells, along with transcriptomic and genomic data, All of this gives us increased confidence in the relevance of our targets to actually make a meaningful difference in improving the outcomes for patients and having the ability to better understand the potential impact long before we reach the clinic. Once we've established the desired target, we rigorously define our objective. the target product profile, or TPP, which describes in detail the properties we desire in our optimized drug molecule. Once we have rigorously defined the target product profile on slide eight, we now take this set of objectives and encode them as a reward bundle for our algorithms to optimize towards. This enables our design systems to create structures meeting those criteria. For example, we may want to design a brain penetrant drug that has a low human dose, good selectivity, but in particular avoids having efflux issues. We can encode that specific set of objectives, potency, selectivity, efflux, et cetera, so that normal structures generated drive towards these criteria. As you might imagine, we use and generate a lot of data when doing this, illustrated on slide nine. For each project, we generate the initial hit structures algorithmically from integrating any public data with proprietary data from fragment or focus screening, which is developed in-house. As you've heard us talk about, half of our company are drug discovery scientists, generating proprietary assays and data at our Vienna and Oxford labs that we can bring to bear within our projects. In addition to that priority data, the platform can also scour existing data, going back years to search for anything that might be relevant. For example, data extracted from a 20-year-old patent or a recent Nature paper can all be integrated with data generated in our labs this morning to help serve the models. One of the great powers of our AI platform design is that we can use any type of data to drive the design process, meaning it does not require a specific data type like 3D crystal structures or high content images, but we can use any and all of these types of data, plus many others that will enable us to triangulate towards designing drugs that meet complex design requirements. This diversity of data is required to precision engineer a novel chemical series that we anticipate will have a robust treatment effect in patients. In order for our systems to generate potential molecules, we need models to predict all of the properties that we require. This could include potency, acne, selectivity, physical properties, and many, many more. We have extensive model-building capabilities that span the full range of skills and technologies, from quantum mechanics and molecular dynamics to exploit structural information, to machine learning and computer vision to interpret pharmacology and cellular imaging. Going back to our earlier example, where we highlighted that we are trying to design molecules that meet specific project requirements, for example, the right level of selectivity and potency, but that doesn't have unwanted issues. We are now at the stage where we have identified the desired TPP, and we have an initial set of models that will help guide us on that journey, as you can see on slide 10. We can now apply generative design, which is an AI-driven process of molecular ideation. Our system is exploring nearly the entirety of chemical space and creating molecules to meet our desired criteria, scoring them and learning. learning from the scores how to create better molecules. Using evolutionary algorithms or reinforcement learning, the system rapidly and efficiently explores chemical space, creates a population of novel molecules that are predicted to meet our criteria. At the end of each iteration, usually a population of tens to hundreds of thousands of molecules are created. These molecules are driving towards the criteria that we desire. On the next slide, from this large population, we apply a detailed filtering process that may involve more sophisticated and compute-intensive models to reduce the set. And then we apply a process called active learning. We want to make as few molecules as possible because it's time consuming and expensive. Usually, we make 10 to 20 molecules per design cycle. Therefore, we want to synthesize and test the compounds that will help us learn fastest to improve our models and to take us forward towards our objectives. It is by learning faster to navigate across a potentially vast chemical landscape that gives us the industry-leading productivity metrics that we have been demonstrating. Our active learning algorithms ask which molecules will provide us with the most information to improve our models in a certain dimension. In short, what should we do next in order to learn the most and to select this set of molecules in an unbiased and mathematically rigorous way so that they enable us to learn the most at each cycle? Now on slide 12, the selected molecules are synthesized and tested. We profile each molecule in detail so that we can update our models with new information and learn the maximum amount from the laboratory work. We have extensive biology capabilities in our labs in Oxford, including structural biology, biophysics, and pharmacology screening. We can then visualize the project telemetry, the progress of the project, in an unbiased way using what we call a merit score as a representation of the desired target product profile, as you can see on slide 13. Each dot is a novel compound synthesized and tested. The x-axis is the sequential progress of the project in terms of compound numbers, and the y-axis is the multi-parameter optimization score with one being the ideal score across multiple objectives. Each design cycle is colored from red through to blue. As the project progresses, the system moves from exploration, where we are exploring a range of different chemotypes. Once the most promising series is identified, we move into an exploitation phase, focusing on a particular area of chemical space. At this stage, molecules are consistently fulfilling most of the key project goals and will rapidly close down on a candidate molecule suitable for preclinical testing. As we learn through each cycle, we can track the learning as the project progresses towards its desired criteria. On the next slide, you can see that the AI algorithms are refining the final designs in order to achieve the project's potency, selectivity, bioavailability, and safety requirements in a final candidate molecule. Hopefully, I've shown you how we design differentiated molecules. It's one aspect of our end-to-end platform. On slide 15 is our learning loop. By starting and ending with the patient we can apply the platform to produce new candidate medicines with attributes that we predict will lead to better treatment benefits. We are also using our precision medicine platform in biomarker discovery and in patient stratification as we move forward. You will hear more about this later in the year. So there's no better way to showcase the true value of our design capabilities than with an example.