The building blocks of AI success stories

Dr Mike Tarselli, Chief Scientific & Knowledge Officer at TetraScience, was Scientific Director of SLAS until 2020, and will be chairing his first event this year. He shares his insight with Megan Thomas. From pharma to non-profit “I probably wouldn’t have the role I have now if I hadn’t worked for SLAS,” Tarselli reflects. When he first got involved with the organisation, he had been the Associate Director of Novartis for the previous five years. What followed was his move from a large pharmaceutical conglomerate to a non-profit organisation. He describes this as being thrust from a solipsistic, internal set-up, where Novartis would essentially do what it wanted with its data systems and science, into a much broader canvas with all the leaders of third-party companies and software companies, instrument companies, pharma companies... and being able to talk with him at a moment's notice. “If I hadn't had that,” he says, “I would certainly not be working at Tetra, today.” AI in drug discovery Answering the question of where technology is being implemented to progress the drug discovery process, Tarselli says: “You know that what everyone wants to say right now is AI (artificial intelligence). However, I'm going to offer a contrasting opinion: We want to get to AI. Everyone wants to use AI. However, we’re still a couple of years early for general purpose AI in science. Why? Because first, we do not have standardised processes across the industry. Every company does their purification differently, every company does their sequencing slightly differently, every company does their assays slightly differently. It's not fully standardised, so companies still do some boutique work. In addition to that, we don't have consistent data standards, or mature ontologies.” Tarselli explains that for AI to function, it needs large, consistent, clean standardised data. “We can do AI on things like facial recognition, driverless cars, and detection of purchasing habits because there are clean datasets that are in a standard format that are exchanged with trillions of data points.” He adds: “Biopharma certainly has trillions of data points, but they're just in so many different formats, places, types, in people's heads and in USB drives that we can’t quite get to general AI, yet.” He continues to say that the next two or three years will be all about getting what the industry collective knows into a form where we can actually learn from it. He says: “Then, maybe in 2025/2026, we're ready to apply vast and dramatic AI models, because then we'll have it.” The building blocks for AI success So, what needs to happen to ensure that AI is a technological success story in the future? Tarselli answers using his SLAS programming as a backdrop. He says: “My track is called Data Science and AI, and it's there as a way to get people attracted to the session, but only one of those three sessions will actually be about AI, directly. The other two are going to be about precursor activities.” The first session – or step to the end goal of successfully implementing AI – is about building out an infrastructure and asking how you get the things 'underneath' to work. Tarselli says: “We need to get the systems built, get the data in place, get the data upgraded, etc.” The second session covers machine learning (ML) models. Tarselli continues: “So, now we need to build the plan. What is it we're going to find with this data? What is it we're going to detect? What matters, what doesn't? What are the weights? How do you know scientifically what's valid and what's not? Then, and only then, can we get to that third session, which is AI Success Stories.” Tarselli says we are in these early days when it comes to AI success stories, and that only a couple companies have done it right. This session will ask, have you used AI and ML to generate actual outcomes for a drug process? For a patient? For a batch release process? Where has that actually happened today? Popular press will suggest that everyone's doing it; there are maybe two or three good success stories per company in 2023.” Technological advancements According to Tarselli, the technological influences we’ll see in the immediate future include the data cleansing and a shift towards consistent automation. He says: "I'm certainly not arguing for taking away scientists’ jobs. I'm simply saying that in nearly every other industry, in retail, in fast food, in banking, we automate processes so that humans don't have to do boring, pedantic work. I'm here to tell you that things like pipetting, screening a plate, moving cells to assay plates... machines can do these now. They can do it really well and consistently, and produce good data. For that reason, the whole society of science has to move towards robots, automation, procedural automation, and data automation wherever they can.” He continues: “It's not like we're going to fire the old pipettors! What we'll do is we'll have them train the robots, look over the data sets, or they can do something even cooler. Or, they can tell us what the next thing we should be doing in this space is. You free up lots of human capital this way. Everybody focuses on what more can be done, but we should also focus on what can be done better. For instance, how do you eliminate mistakes? How do you decrease risk? How do you increase compliance? How do you decrease expenses?” Uniqueness of SLAS events Tarselli says that there is a thing you can do at SLAS that you can't do in many other places, which is that you can stop thinking only about what your company needs and you can consider those needs in the context of the community and the field, and look at how to do work differently based on what others are doing. He says: "There's going to be 6000 to 7000 people at this conference, and they're all going to be able to share opinions – some will be bad and some will be good. You get the opportunity to not only know what you should do, but also see where mistakes have been made, and where to avoid errors. There's just endless learning." What to expect from 2024 Looking forward, Tarselli says: “If you look at buzzwords in search engine optimisation, of course you will see AI. However, behind the scenes, if you pull back that one layer, CRISPR is everywhere. Ever since the 2020 Nobel Prize was awarded, it's just gone bonkers: new start-ups, new cell therapy techniques, new assays... All the best attended sessions at SLAS 2023 were CRISPR-based. Recently, the very first human in vivo CRISPR therapeutic was approved in the EU for sickle cell disease, by Vertex and by CRISPR Therapeutics. I know everybody wants to do AI, and I get it, but there are the ‘pots and pans’ things you have to do to get the work done, and CRISPR meets that need. It leads you to new assays, that leads you to new modalities, leads you to new tissue types, leads you to new organoids. Being able to dial in the genes you want is amazing, and we've never had that layer of control. When I was coming up in the world, we had cellular assays, bombarding things with giant soups of chemicals, and that was the best you could do. Maybe you could get an antibody in there. But now, you can literally open up a cell and say what you want it to do.” In addition to CRISPR, which Tarselli refers to as his 'dark horse', he also highlights the importance of ontologies, without which you can’t interrelate what you know. Without ontologies, he says you can't, for instance, learn something from an assay and notice that it corresponds to a drug, or to a target or patient phenotype. He says: “Unless you have a way of relating terms and saying ‘this is this, this is not this’, how are you going to take in all that data? You're formalising across trillions of data points and relating it to something you know. So, you're going to see a lot of ontological characterisation and discussion at the show.” SLAS 2024 Supplement, Volume 25 – Issue 1, Winter 2023/2024 Biography Mike Tarselli, PhD, MBA, is Chief Scientific & Knowledge Officer and Head of Quality & Compliance at TetraScience. Tarselli expands scientific applications of the Tetra Scientific Data Cloud. Previously, he was the Scientific Director for SLAS and an Associate Director at Novartis. Tarselli received his PhD from UNC Chapel Hill, completed postdoc at Scripps Research, and his MBA through Quantic School of Business & Technology.