Can AI Cut Medicine Costs? The Future of Pharmaceutical Manufacturing | Martin Wood & Adrian La Porta

As AI promises to accelerate drug development, a critical question emerges: can pharmaceutical manufacturing keep pace? In this episode, Martin Wood and Adrian La Porta examine how artificial intelligence could transform the entire pharmaceutical value chain, from quality control paradigms to facility design.

Recorded as a preview to Bryden Wood's Accelerate Pharmaceuticals event on 25th November, the discussion explores fundamental questions about the future of drug manufacturing. Adrian argues that current quality systems are essentially compensations for our limited understanding of manufacturing processes - and that AI could enable a complete reversal of this paradigm through real-time, data-driven quality monitoring.

The conversation ranges from the potential for patient data feedback loops that could reshape manufacturing in near real-time, to whether facilities should be designed around autonomous systems rather than people. They examine why continuous manufacturing has struggled to gain traction, whether digital twins could eliminate traditional scale-up challenges, and how distributed manufacturing models might reshape global supply chains.

A thought-provoking exploration of whether AI can finally bridge the longstanding gap between drug development innovation and manufacturing industrialisation - and what it will take to get there.

To learn more about Bryden Wood's Design to Value philosophy, visit www.brydenwood.com. You can also follow Bryden Wood on LinkedIn.

Transcript

martin wood: 0:00

Is ai, the initiator for, you know, a real next generation of drugs. Discovery and drugs development and drugs. Drugs, manufactured

Adrian La Porta: 0:09

quality systems are in some way a compensation for our lack of understanding of what's going on in the process.

martin wood: 0:16

That's probably one of the areas which has one of the best opportunities to harness. The power of artificial intelligence going forward,

Adrian La Porta: 0:22

regulation is about trust. At the end of the day. Very human quality at the moment. There's no feedback loop.

martin wood: 0:28

Part of the problem is people not getting together. Hello, uh, my name's Martin Wood. I'm here today with Adrian Porter, uh, to talk about an event we have coming up an accelerate event at 1 0 1 Houston Road on the 25th of November, um, which is about the role of artificial intelligence in drug manufacturing. Um, we've got a stellar cast of people coming to discuss the subject. Um. And really that's a reflection of the fact that it's, uh, an interest to so many people, uh, because it's, uh. Subject close to our hearts, which is how we, um, what the rate limiting step that manufacturing presents in the introduction of drugs to, um, you know, the good of society. Um, so we're gonna look at all the changes, uh, that might be necessary. Um, the differences in the way we think about it, uh, that we're gonna be required if this is actually gonna be, we're gonna accelerate this process. Um. For the benefit of all. Um, so to that end, I'm going to, uh, ask Adrian, uh, a few questions around this subject because, uh, this is a subject that, uh, um, I've been, uh, involved in this for probably 15 years, but, uh, Adrian much longer. It'll be 25 years in this subject area. So, um, process and design and optimization. So one of the questions is what can, what, what, how can we bridge the gap that's, uh, traditionally. Existed between, um, the drug development, the drug development process, and the industrialization of those drugs. So we, I think, believe that drug development is gonna accelerate by the use of artificial intelligence. Mm-hmm. There's been a lot of investments. Some of it hasn't come to fruition yet. It's been a number of articles about this, but, um, but it's coming. And they will inevitably have an acceleration. Now, of course, that's a fantastic thing, unless we actually can't respond to it from a manufacturing perspective. Um, what are your thoughts on that? A

Adrian La Porta: 2:30

uh, well, I guess I'll start by talking about the event a bit more because, uh, um. We have, um, people from, uh, big tech, from big pharma, from the, uh, the biotech, ai, uh, sector as well, both from academia, uh, and, uh, innovative companies in the knowledge quarter. And, um. The reason I mention that is because, you know, you know, we're trying to find out the answers to those sort of questions at the event. So, uh, we're, we're, we're very much, I'm very much looking forward to discovering more from that, uh, assembled, um, knowledge and, uh, and expertise we're gonna get on the day. Um, so the, um, the premise for the event, as you said was, um. There's an expectation that AI in drug development is going to feed more new molecules, new products, into the manufacturing pipeline, uh, which is the area we're more familiar with. Um. Um, it's not clear to us at the moment how manufacturing is gonna cope with an increased, uh, number, uh, rate or, or level of sophistication of, of drugs, uh, coming into, um, coming from this AI fueled, uh, development pipeline.

martin wood: 3:55

I think the right at the heart of that is, uh, is how. Um, quality evolves. So the, the whole subject of quality in drugs manufacture, um, how that evolves. And, and that's probably one of the areas which, um, has one of the best opportunities to harness the power of artificial intelligence going forward. I suspect.

Adrian La Porta: 4:15

Yeah, I, I think it's widely acknowledged that life sciences hasn't. Uh, graft, um, AI in certainly manufacturing, uh, as quickly as some other sectors, and that's understandable because of the highly regulated environment that we're, we're used to working in. Um, but um. As you say, the, the potential for, um, AI and advanced control systems to actually improve quality. Um, I is clear. Um, perhaps the, the challenges around the ways of working, the paradigms, the culture that we're all used to, uh, within manufacturing. But

martin wood: 4:59

there's a bit of a conundrum in the sense that. Um, regulatory frameworks and, uh, quality frameworks are quite static. Mm-hmm. And this is obviously a very dynamic situation and getting increasingly so.

Adrian La Porta: 5:10

I, um, I think you can look at this, uh, at a, at a couple of levels. I think, um, at the, at the, at the manufacturing level, at a, at a more detailed level, we've, we've all got used to the idea of quality being about. Making things static and locked in and um, uh. Regular and repeatable. And that's, that's the paradigm we working. But, and there's another way of looking at it, perhaps where you go, well actually to, to, to know that a particular batch of a, uh, drug product is good quality. Um, so in, in this paradigm I've just talked about, we say, well, we, you know, we always do it the same way. We measure the same parameters. Um, we take a sample, we, we run it through a standard method. It's all about standardization and, and, um, uh, keeping things the same. So, but in reality, our understanding of what's going on in the process and what's happening in that tank is actually very poor because we're looking at a few data points and, uh, our, uh, the way we lock things down is, is. An antidote to our lack of understanding of what's going on, you know, in the equipment, in the process. So if you can look at quality perhaps in a completely different way and go, well, if I really knew what was going on, I would know exactly the quality of every step in the process. From having vast amounts of data, which we can gather from the, from the plant, and really correlate. Um, uh, quality with performance in real time. It's a completely different way of looking at quality, and, and I think it's actually a better way of looking at it. Uh, but it, it, it requires a complete, you know, reversal of our, of our mindset. I mean, I

martin wood: 7:01

guess that starts really in the sense that in drug development, because we've always, uh, we've always. Found the gulf between drug development and industrialization. Yeah. Um, sort of an interesting area for potential change and improvement. Um, I mean, the introduction of continuous small molecule drugs manufacturer in terms of, uh, first intent by drug development was maybe 20, 25 years ago. Um, and yet we still don't see more than really a pilot plant. Um, pilot plants using continuous processing. Um, and that's, uh, largely because it was almost, I mean, it wasn't as simple as this, but it was almost like someone didn't tell manufacturing it was coming. Um, slightly flippant, but um, but it wasn't ready, uh, really for, for how the drugs were being developed. And, and, and if, if it had been then it probably would've led to, uh, higher efficiencies already. So. On that basis, understanding how much monitoring there can be of a process and how dynamic that can be as opposed to, uh, locking in a static recipe effectively. Mm-hmm.

Adrian La Porta: 8:11

Yeah, and I think the continuous manufacturing is a good example.'cause it's a, it, it, uh, for, for continuous manufacturing to work, you have to have a better understanding of how your process works, which is perhaps why it's, you know. Been so resisted, you know, as, as going back to what you're saying, the, the quality systems are in some way a compensation for our lack of understanding of what's going on in the process. And when we try to move to continuous, you know, we have to do more. Investigation. We have to understand how things work better. And you can extend that, that thinking to the whole of manufacturing. We, we've seen an example recently, um, uh, on a project for instance in um, um, inspection. Uh, you know, final inspection of sterile products and, uh, um, that, that's really interesting.'cause it's, it's a, um, a, a, an AI powered system that's going into manufacturing in, in commercial now, it's increasing yield, uh, and increasing quality. So it shows, shows that it can be done.

martin wood: 9:13

Well, that's quite interesting because that's, that's, that's, uh, augmentation of an existing system.

Adrian La Porta: 9:17

Yeah. Uh, which is a

martin wood: 9:18

great thing. Yeah. But one of the questions is, is the existing system quite fit for purpose? Controversial, uh, discussion. But, uh, maybe if there was a, some changes to the, uh, basic framework structure of, of, uh, quality regulation, then it would stop, uh, only being able to be implemented in pocket areas where it's augmenting, uh, just augmenting the, uh, quality.

Adrian La Porta: 9:44

Yeah. Uh, I, I, I said earlier, there's maybe two levels of looking at this and, um, I was talking about the manufacturing level and, and the, the, there's the higher level, the more systemic level is thinking about how, um. How AI might change the relationship between manufacturing, development and, um, use of, and medicine, if you like, use of the drugs, uh, in the population because we might see, uh, an explosion in the amount of data that we get from people using the drugs that we're manufacturing. And at the moment, there's no feedback loop, or the feedback loop is extremely slow. So, you know, we're strange

martin wood: 10:29

in this day and age of wearable technology.

Adrian La Porta: 10:31

Well, exactly wearable tech, you know, diagnostics. Uh, if we at the moment what we, you know, the, we do a clinical trial on a limited population, we look at the results, and then the drug goes out into the, into the world. There is post-launch monitoring. But, uh, you know, you could see, uh, or maybe the challenge. Is how do you use that data in the future to, uh, feedback into manufacturing in, in, you know, not quite real time maybe, but, but on much faster loose. I mean, and

martin wood: 11:03

these are the kind of things that we're hoping to, um, sort of begin to hypothesize. Yeah, there are no answers to these things yet, but, uh, the point is. Um, there's a whole range of people that don't necessarily normally coincide in their normal lives required to make those kind of things happen. And that's, yeah, kind of what this event has, uh, has is its strength, is it's got a number of different areas, as you said, from big tech, um, through to drugs, manufacture being represented. So, uh, those are the, some of the, um, more radical things that possibly we could hypothesize. Um, the other thing about it, of course, is that, you know, we. Drugs manufacture at the moment is quite a, quite a, um, personnel intensive process. Quite a, quite a significant amount of labor and maintenance and equipment, um, reliability issues in the process. And of course, again. I think one of the things that's gonna come through here is the more data you have and the more monitoring you do, the more, the more you can affect this process, the more artificial intelligence can, can actually start to, um, uh, start to improve those points. But it's not just about the drug itself, it's also about the equipment and the facilities and so on, which they exist. Um, making that a, uh, modifying those, um, feedback loops so that we get a much, um, we get a much more, a much higher availability. Uh, we get better OEE from our processes. Mm-hmm. Which is extremely important given the kind of investment that goes into these facilities in the first place.

Adrian La Porta: 12:33

One of the themes here, I think is the way that the, uh, uh, the use of, of these type of, um, systems starts to break down. Um. The, the barriers between different parts of the, the whole development, manufacturing use sequence, um, and uh, uh. Make the whole thing more, uh, integrated. But so, so one of the things we, uh, uh, I I, is how do you, when you transfer, uh, a product from chemical development or through the development process into manufacturing, uh, that we, we, we haven't spoken about is the use of, um, uh, simulation and digital twins. So, you know how your drug's gonna scale up, you know how your drugs are gonna perform in the factory. Before you ever put, you know, do that physically. So using, uh, uh, uh, that's, it's a bit of an overused term, the digital twin, but being able to predict how things are gonna operate, um, before you do it, uh, in the physical world is gonna be very powerful. Because again, at the moment, the way we do things is when we put things into the plant for the first time, uh, it is kind of guesswork, you know, put it in, see what happens, which makes what makes commissioning exciting. Um, but also once we're up, up and running, as you was saying, um. We have to understand how plants gonna be perform over time. So things like predictive maintenance, real-time monitoring are already there. But you layer on artificial intelligence on top, you'll know exactly when to, to change a, you know, a pump seal or when to, um. Uh, look at modifying the parameters in a control loop. So all that's gonna increase efficiency in turnaround times. And,

martin wood: 14:23

and I guess guess one of the other really, uh, interesting subjects is gonna be, um, automation, robotics, which of course, you know, tell me confused directly with artificial intelligence, but there's obviously gonna be very synergistic. Um. You could argue that a lot of traditional facilities are designed around people Yeah. And the way, and the ergonomics of people. And, um, one has to question whether that's the necessarily the way forward. Um, because actually some of the equipment's been sized and the whole of the process has been sized around, you know, dispensing things into a vessel levels of a building where you. Can reach parts of the process when in actual fact many of those things could be more autonomous, they could be more, um, uh, robotically, uh, driven material handling, uh, processes within. And that may in fact, uh, change the paradigm with regards to the whole scale of manufacturing in the first place. Because

Adrian La Porta: 15:22

yeah, one

martin wood: 15:22

of the issues with manufacturer is that, you know, generally. Speaking, it has to have to suffer. You referred to it as the excitement of commissioning. It's had to, uh, suffer the scaling up. Uh, process from, from, at least from a pilot plant, which, uh, is, uh, fraught with potential, um, difficulties and, uh, could cause, uh, you know, could cause a, a, a drug to fail. Uh, it may change one's whole approach to the, um, speed of the manufacturer. Uh, speed of the manufacturer and, uh, if you could automate the material handling characteristics and break the, this sort of idea that, um, this is a batch by batch, uh, person, material handed, uh, handled kind of process.

Adrian La Porta: 16:10

Yeah, I, I think there's a, um, obviously, uh, automation, uh. Is useful and automation's getting more reliable. Um, but I think it's also a mistake to try and take a, the, the sort of human-centric process and just automate it, you know, sort of ult. The ultimate end of that is, you know, uh, uh, people, you know, basically, um, just picking up the ke and pouring it in. So, um. But the, uh, tying it back to what we were saying earlier, part of the reason that we have batch manufacturing of a certain size is the idea of the static control of quality. So if we look at, um, small molecule, for instance, we have a paradigm where we, we do a chemical synthesis step and we isolate it as a dry powder. And that's partly because we don't understand what would happen or we're worried how complex it is if you just take the synthesis from step to step without isolating it. So your whole architecture of, of manufacturing is driven by this really this, this idea of having to have, uh, you know, batches and isolation at a size that people can deal with, you know, at, at a rate of. Analysis that people can take something off to the lab and read a result and put it in their log log on their, the lab book. Um, so all that goes away potentially if you are able to extract data, analyze it, and with high degrees of assurance, understand what's going on in the process. You say. Interesting. Yeah. Much more. You mentioned people many times, right? People centric. Yeah.

martin wood: 17:45

Um, one of the issues there really is that, um, I think that's one of the things that reassures us the. Call general public is that it is so people centric. But the question is, is that right way? Well, it's a very point in the future. So, so, you know, you could hypothesize that this, that the manufacturers of drugs should be a less people centric process. And what are the implications of that?

Adrian La Porta: 18:09

Yeah. I suppose regulation is about trust at the end of the day, and the people are taking very human quality, are taking, are taking drugs. They wanna know that they are. Yeah, absolutely. They're safe. And we, there

martin wood: 18:20

is a very deep philosophical Yeah. Um, issue that this will, uh, that artificial intelligence I think will provoke, uh, within people. And I think we need to consider that in the background. Yeah. If we consider these more,

Adrian La Porta: 18:33

I imagine it's a. You know, a discussion that comes up in the application of, uh, artificial intelligence in every field. Well, it does. Does, at what point do you lose? It does just particularly

martin wood: 18:43

pointed

Adrian La Porta: 18:44

when it comes

martin wood: 18:44

to,

Adrian La Porta: 18:44

so we have to think about. AI as a human enhancement. Yeah. And, and not to get to the point, well I pressed the button, the drug came out the other end, but I have no idea what happened in between. No, absolutely. That's not really where we wanna end up.

martin wood: 18:59

One thing that it should be able to do seamlessly, and as we talked about some of the different parties involved in the accelerate, um, is bring some of the areas of, um. Well, the pharmaceutical companies where they operate, bring them together, uh, in a more continuous digital process. So, um, the actual supply chain and also even, um, the strategic, uh, decisions that are made upon, um, the investments into different drugs, um, are, are somewhat disconnected traditionally. Mm-hmm. Um. They're not necessarily, uh, holistic processes where they're all considered as, uh, um, from, you know, cradle to grave. Yeah. Simultaneously. And I think the, one of the things that artificial intelligence, I mean just the whole digital world, but very much more artificial intelligence could help with, with, is, um, creating this, um. Much more, uh, end-to-end integrated process from, from strategy through to supply chain, through to manufacturing, um, and, and through to healthcare. In fact, into the healthcare environment. Exactly. So it's a question of how does that, does that need a new kind of backbone to it effectively is one of the questions I suppose we'd like to consider.

Adrian La Porta: 20:18

Yeah. I was, uh, listening to something about, uh, some of the vaccine development and, uh. It was, uh, uh, this was, uh, Emma Warley. GSK was talking about this, about how, um, you know, using large amounts of data with the, the help of AI could help, uh, target clinical trials to, to much more precise populations. Um, which means that you can, um. Your chances of hitting, of getting a successful clinical trial, uh, are higher because you've targeted the right population, which isn't cheating. It's just finding the right drug for the right people. Yeah. Um, you know, it's, and it sounds like it from a's perspective, but it's

martin wood: 21:04

in fact it's not.

Adrian La Porta: 21:04

Um, it's sort of the other side of that, that thing we've talked about for a long time about personalization medicine. You know, why, why, why? Put somebody into a clinical trial that you've, uh, through analysis found out, wouldn't benefit from that drug anyway. It doesn't make sense. But, um, it is a case of integration of sort of the healthcare monitoring with the drug development process. If you think about influenza. Um, that's a sort of extreme example because every year you have to make a new product, you know, it's a Yes. And people would be quite familiar with that. And that's, and you have

martin wood: 21:43

to have, uh, large areas of drug plants laying dormant until, until they might have to, uh, increase demand, uh, for, for drug overnight for sure. And it's not that effective either. Well, and it's incredibly poor, it's incredibly poor financial model for pharmaceutical companies. Yeah. Because, um, who wants that kind of peaky manufacturing?

Adrian La Porta: 22:03

So I suppose what we're saying is that there's this ability to deal with very large amounts of data in. Relatively short periods of time opens up the possibility of a much more effective, targeted, um, uh, pharmaceutical industry.

martin wood: 22:21

I mentioned it earlier, but do you think this will generally enhance the opportunity for continuous manufacturing to become a bigger player in the drugs, um, uh, synthesis of drugs in the future because, um, it has, uh, had a somewhat. A full start really in manufacturing and uh, and I think that's largely because it's been, it's come across the, some of the paradigms we talked about before, which is the people-centric nature of it, of the manufacturer. Thus far the scale of manufacturing being related to people, which continuous, continuous manufacturer doesn't have anything to do. I think the

Adrian La Porta: 22:56

short answer is yes, but I would generalize it into more intensified or more efficient. Manufacturing. Yeah. I

martin wood: 23:03

mean, accepting the fact that many stages still will never be able to be continuously

Adrian La Porta: 23:07

Yeah. It might not be the right, uh, right thing. So one of the models we'd looked at, uh, was, uh, scaling things down to the point where, um, you've, you've effectively got a, a robotic laboratory scale process because some of these, uh, uh, high potency drugs, the amount that you need. Uh, is only in the tens of kilos per year. So I mean,

martin wood: 23:29

obviously a, a byproduct that might be the, um, smaller facility where, where you can place the facility in a, let's say. A, uh, politically better network strategy because there obviously is challenges going forward in terms of the global supply chain.

Adrian La Porta: 23:45

Yeah.

martin wood: 23:45

And, um, wish for onshoring and so on, and certain drugs. Um, it might be that, uh, facilities can actually be rather more agile and be, um, perhaps smaller and perhaps more, uh, population targeted than they are. Yeah.

Adrian La Porta: 23:59

Yeah. You can certainly go for a distributed manufacturing model. Uh. And again, coming back to, to this theme of breaking down barriers between different stages. If you end up with a manufacturing process, which is at lab scale and distributed, how much difference is there between that and the, uh, development process anyway? You know, are they, are they gonna merge? Uh, and if you are taking real time data from the, from the patient population and altering your drug, um, in response to that, which admit is very, very, very out there compared to our current system, then, um, uh, development and manufacturing start to look like the same thing.

martin wood: 24:42

Mm-hmm. Interesting. Um, so one of the things that we, uh, I guess one of the things is. Is ai, the initiator for, you know, a real next generation of drugs, uh, discovery and, uh, drugs development and drugs, uh, drugs, manufacturing is, is it, are we gonna get, are we gonna be able to put together a, a, a new era of drugs manufacturing? This is, this, is this kind of what we're looking for because it's quite interesting. Some of these initiatives existed before and won. It springs to my mind, is what they call fourth generation. Um, small modular nuclear act is very popular in public consciousness. But actually what actually happened, uh, um, was um, investment happened in an incredibly, um, dispersed, diverse manner. And you ended up with literally hundreds of different, um, potential, uh, solutions to the same problem, which is great in one sense, but it rather diluted the investment. Uh, into new nuclear and made it, um, one might argue, um, didn't forward the cause of it as well as it possibly could. So part of the problem is people not getting together. And, uh, finding a kind of platform or a backbone or, or a, um, sufficiently coherent, I suppose, approach where what they would be, what each party would do, would reinforce the outcome of the other. Um,

Adrian La Porta: 26:13

yeah,

martin wood: 26:14

so I, I wonder whether that's in the world, which is as competitive as pharmaceutical, whether it's possible actually. Um, to get at least the knowledge, the backbone and the knowledge to be, um, rather more focused than it might be, and therefore the investment to be rather more, uh, targeted and, um, effective than it might otherwise be.

Adrian La Porta: 26:36

I think there's a balance, isn't there?'cause coherence is the enemy of diversity and uh, you know, you can, you, you, you're in at risk of backing, you know, picking winners. Yes. But, but I think there's plenty of examples of where things like common standards working together with it's fantastic regulations, fantastic dilemma, isn't

martin wood: 26:55

it really? Yeah. It's fantastic dilemma that picking winners. Uh, um, but, but at the same time then, then just. Making it hard to disperse the effort.

Adrian La Porta: 27:04

So, so, but, but I think there's plenty of scope, if you like, for, um, areas where, um, drug manufacturers, medicine companies, healthcare and regulators would on wanna work together to, um, establish. Common standards. Uh, and this happens all the time, doesn't it, in in regulated environments, uh, whilst allowing, you know, the, the, the innovators to, to work within a framework. Um, and again, there's another spectrum of, of, uh, uh, you make it too restrictive and you. You, um, limit innovation. If you make it too lax, then uh, you lose public trust in the, in, in, you know, in drug regulations. Well,

martin wood: 27:47

yeah, absolutely. And the public trust, as we mentioned before, the trust word, um, you know, AI promotes suspicion from that

Adrian La Porta: 27:56

perspective. Yeah.

martin wood: 27:57

From the offset. So, um, so I guess it has to be, I mean, again, reason for having an event like this is to, is to, to, to get that thinking started and to get it to be coherent and to get it to cross the boundaries that don't normally get crossed.

Adrian La Porta: 28:11

Absolutely. And so, so maybe that's, uh, you know, will, will, I'm very much looking forward to the event and, uh, absolutely hope to, uh, learn a lot more. And, and as you can about the question asked. Well, as you can hear from this, we've

martin wood: 28:22

got a lot more questions than we have answers at the moment. So, um, what we'd really like to do is to at least start, uh, engender more, um, uh, um, combined critical thinking. Across, across a broad group of people and see if we can get, um, some hypothesis of, uh, what, how we one might go forward. And, um, and then if nothing else, just promote those kind of, um, collaboration that, uh, that, that might not already exist and needs to exist in the future.