00:00:04: The Biorevolution Podcast.

00:00:06: Your hosts,

00:00:07: Luise von Stechhoff

00:00:08: and Andreas Roichler.

00:00:11: from animals to algorithms how AI brings drug testing closer to human biology.

00:00:18: this episode's title of the biorevolutions podcast welcome easy.

00:00:22: so today without a guest because we have an intense dialogue.

00:00:26: let see how intense it is going.

00:00:30: Ethics involved, of course.

00:00:32: But practicalities as well and maybe groundbreaking new science when it comes to replacing animal testing.

00:00:41: all that in this episode.

00:00:43: but As always we start with quotes.

00:00:45: What do you bring?

00:00:46: I brought a couple of quotes because i think It's pretty broad topic That were discussing today.

00:00:50: so how can We replace reduced animal testing with the use of artificial intelligence.

00:00:56: One of the quotes comes from Jama Davis, from Animal Free Research UK.

00:01:00: She was quoted in a recent article that came out just the other week.

00:01:04: so we're February.

00:01:07: What she says is experimenting on animals causes pain and suffering.

00:01:11: This becomes an especially hard pill to swallow when you realize that in many cases the experiments do not translate into benefits for humans, And this I think it's a basic premise of what we're going to discuss today.

00:01:22: When does it really warranted?

00:01:24: To use Animals In testing drugs and testing hypothesis For humans if they are Not really translatable.

00:01:32: This Is what We're Going to get Into.

00:01:34: The other Part Comes from Karen Blith or Karen Blythe, quoted in the same article.

00:01:39: She's a professor of In vivo cancer biology at The Cancer Research UK Scotland Institute in Glasgow and she says our options for fully understanding metabolism and anything other than living mammalian model are still very limited again something we're going to get into.

00:01:56: if we replace models how can we really capture physiology in holistic fashion?

00:02:02: And the last one I want to quote is not a person, but an agency.

00:02:07: The FDA in the United States who were in recent road map or two phase out animal testing for specific drug classes says In the long term three to five years.

00:02:18: that long-term FTA will aim To make animals studies the exception rather than the norm For preclinical safety and toxicity testing which Is pretty bold statement Absolutely.

00:02:30: If this is realistic, I think it's what we want to discuss today?

00:02:33: Yeah absolutely so.

00:02:35: To know the lay of the land It does make an awful other sense to dig into history a little bit because animal testing hasn't been something that has been around forever but for quite long time.

00:02:47: and How did these all get going?

00:02:50: Animal testing has been Around for awhile since humans have discovered That they want to study biology first of all, ethically possible to do that in humans and many cases.

00:03:05: And it's also not always... feasible to do this in models and then studying, for example behaviors but also studying responses to certain compounds in animals was seen as a proxy.

00:03:16: To better understand what would happen any human situation?

00:03:20: And I mean we talk about animal research now In the very broad sense.

00:03:24: there are many different ways of animal research.

00:03:26: Of course There's academic research that really studies only like fundamental biology Really trying to understand in a translatable fashion for humans, but also sometimes really just understanding how the certain type of animal would behave.

00:03:42: What kind of physiology this animal has and so on?

00:03:46: But there's also the type of animals research that is done mostly compound, a certain therapeutic is safe for humans to use.

00:04:02: So we want to find out before we move it into the clinical study and test that in humans there will be no harm in this human.

00:04:10: The other thing was also wanting at least an extent certainty of any effect because otherwise it would have been an experiment wasted or ethically more questionable.

00:04:22: That wasn't always In general the idea that you have a regulation of which drugs, You can give to people.

00:04:30: That is well not rather new concept but about like hundred years ago... ...that was not really common.

00:04:35: it wasn't regulated in that sense.. ..you could just sell anything and say this works?

00:04:41: And its safe.

00:04:42: good luck with it!

00:04:43: And I think a really big shift came with the thalidomide scandal.

00:04:47: This is something in Germany that we know as Contagane, it was one of the most famous drug safety scandals in the twentieth century where compounded were sold as cure or treatment for mourning sickness and pregnant women actually turned out to be embryotoxic or heterogenic causing... It's only embryo toxic not heterogenic but it caused birth defects in thousands of children actually instigated this idea that things need to be tested thoroughly before they are given two people and then there should also be tested fairly in animals, before the enter clinical studies.

00:05:25: And here is these two species' ideas.

00:05:27: you test at least for two species – mostly a rodent mouse or rat…and often slightly bigger animal whose physiology is closer pigs, mini-pigs.

00:05:42: For example.

00:05:43: if we talk about biologics so antibodies gene therapies and the like then often they're also tested in monkeys non human primates because certain features for example the immune system that can be triggered by these types of compounds is very different for example in rodents and also not close enough in other species like dogs or pigs.

00:06:05: So we need a model even closer to us to predict this kind of toxicity.

00:06:10: So historically speaking, there is a good reason protecting humans from harmful effects of drugs and also only so I mean safety.

00:06:20: It's the one thing.

00:06:21: The other things that we want to have rather good hypothesis That something actually works especially if you think about for example cancer.

00:06:29: If would test new treatment in patients We want at least an inkling that it will have a benefit for them.

00:06:37: Otherwise, it would really not make sense to give people who we want to cure of disease and new type

00:06:43: treatment.".

00:06:44: Even the quotes at very beginning hinted too this side where in cases doesn't makes one hundred percent.

00:06:55: In many cases, it makes no sense at all.

00:06:58: So

00:06:59: mice have a very different metabolism and they are very different brain with an immune system that has the same behavior as life span.

00:07:08: Mice don't naturally develop many of these types of cancers we have but for sure they do not develop neurodegenerative diseases or type II diabetes in this way.

00:07:21: so if you want to study If something is efficacious in these models, we need to artificially induce this kind of diseases in the mice.

00:07:29: And for cancer, it's very common and also a very common saying if you look at biotech investor decks they have really great response rates then people are like yeah great cured cancer on mice because I mean your artificially implanted tumor often not even from patient derived cell line but Like a laboratory cell line, which has its own flaws.

00:07:53: Let's put it like that You implant that in the mouse whose immune system needs to be reduced To even accept this.

00:08:00: so again you make the situation further artificial.

00:08:04: This is of course very far from and naturally developing disease in humans.

00:08:09: That said there are also mouse models for developing cancer in a more, let's say natural disease course.

00:08:16: But this takes a lot of time and makes it more expensive—makes it less reliable —and also less likely for drugs to show an effect.

00:08:24: so I think there is a little bit of a vicious circle of wanting-to see results?

00:08:30: It's not something that i'm criticizing...it's just kind of the way this type of experiment works….

00:08:36: And the question really how well does this translate?

00:08:40: actually doesn't translate as well.

00:08:42: And cancer is still something that has, let's say a higher chance of translating I think than for example anything related to neurodegenerative or psychiatric diseases because mice just don't get depressed Or not in the same...I can even imagine they do get depressed but i think it's very different way.

00:09:04: then humans are testing.

00:09:05: Maybe

00:09:05: there no such thing as dementia or Alzheimer's?

00:09:09: definitely not in the same way, so I think that learnings we can take away about a disease and treating it from these animal models is very limited.

00:09:20: Therefore here you really need to ask this question which people are doing of course already for long time?

00:09:26: how much do we want to even rely on these models that don't tell us anything.

00:09:30: Because I think there are two things, and what you said in the beginning it's about ethics.

00:09:34: so we want of course spare animal life as much possible.

00:09:38: There is this three R principle replace reduce refine where i mean get rid off the animals as much as possible.

00:09:46: if not limit the number of animals If not try to treat them as good as possible.

00:09:53: really weigh the options of using animals in each and every experiment that you're doing.

00:09:58: And if it doesn't tell you anything useful from an ethical standpoint, I think you should try to avoid any.

00:10:04: But there is another reason which First of all, you spend a lot money on these animal experiments which might maybe from our experiments not be that expensive but for bigger animals and especially for non-human primate studies That are done for antibody development.

00:10:21: This is becomes really really expensive.

00:10:24: So it needs to tell us something meaningful in order to warrant this investment.

00:10:28: I mean, this is one part And the even worst part is what you said?

00:10:32: I mean this translation gap right which we know and we talked about at many times on those podcast.

00:10:39: average industry rate of success from a drug entering clinical development to making it into the market is around ninety percent, sometimes less.

00:10:48: Some other estimates maybe but more depends very much on that type of compound and disease area

00:10:53: etc.,

00:10:54: But you can say almost everything fails.

00:10:57: I think there would be simple way put in And one reason why most things fail because models which make decision Is worth putting something don't tell us enough.

00:11:09: And this is, of course something where we really need to become better because this really wastes a lot of money.

00:11:16: and there's also question of ethics there.

00:11:18: if the industry wastes money on hypothesis that are not prone to play out you deprive patients off better options?

00:11:26: Because I mean in the end what the pharma industry is therefore is to bring innovative treatments for patients to cure disease or to better manage many.

00:11:37: every failed program takes away from other better hypothesis.

00:11:41: So really it's a trade-off to try and make the best decisions possible, on one hand of course to stay financially viable but also in the other hand to follow the mission of getting new drugs out there for patients.

00:11:56: so I think that is really the crux animal testing model that we are relying very much on models.

00:12:04: That might not tell us the right things, and this is really now speaking more for the drug development side of things there as off course.

00:12:11: also the academic research part of Things where you're?

00:12:15: Testing basic hypothesis And they I would say and this also goes a little bit with The second quote that i read so it's A Little With the question what Are We trying to study?

00:12:25: because if we want To Study For example behavior This Becomes Very Hard in model that is not has no body.

00:12:33: So how do you study bodily response to something?

00:12:36: That doesn't have a body, of course You can and I can imagine that this could be Something that you can model that with AI eventually.

00:12:43: but at the current stage AI models are extremely reductionist And they're really studying mostly.

00:12:49: now we're starting To see little bit bigger foundation models that include various parameters But in most cases the models that work well very focused on one single parameter.

00:12:59: this is also the tricky part, right?

00:13:01: Of human biology and disease biology and drug development.

00:13:05: It's very hard to predict what's going to happen because you have heterogeneity between people who have a complex interconnecting system And we test something in like one small answer but would need maybe million answers.

00:13:20: I think there really an opportunity for AI help us finding better and more answers ask the right questions before we put things into clinical development.

00:13:44: The interesting part is that a huge portion, a huge chunk of what you've been explaining now has been known for awhile and industries wish to tackle those problems have been there for while as well And so the approaches are in my understanding.

00:14:01: please correct me if I'm wrong.

00:14:03: In vitro, in silicoe & hybrid

00:14:07: Did you dig

00:14:08: into that a little

00:14:09: bit?

00:14:09: Yeah, I think in silico is anything that... ...is AI algorithms computational.

00:14:14: Anything that doesn't know what lab and then, yeah, in vitro would be anything that's not organismic but in the lab This combination really does make a lot of sense.

00:14:23: so But agencies like FDA are phasing out animal strategy in the UK And on European level.

00:14:29: they all stress these NAMS, the new approach methodologies.

00:14:34: Some people also say non-animal methodologies but I think that first one is more accepted term to use and this are just replacement models.

00:14:44: they work on many different levels so it could really sell experiment.

00:14:48: But there's a lot of sophisticated types coming up Many of them called organs or chips even patients where you grow out Oregon like a liver, kidney something else even hard little brain on the chip.

00:15:06: That's of course not.

00:15:07: I mean you cannot imagine that you have fully formed liver there but you have a piece of tissue that would contain different cell types that interact in the liver and if it is sophisticated model also are hooked up to some type of circulation which comes just through tubing.

00:15:22: so you simulate what the organ would behave like, and then you sometimes even have multiple of these organs linked together with fake circulations.

00:15:32: So a lab

00:15:33: body at the end of the day?

00:15:34: Exactly!

00:15:35: And this is this patient-on-chip idea... I think it's very promising because there are little bit of complexity that we need but also its reductionist enough to be able simple questions.

00:15:50: And it has two huge advantages, and one of them is that you can use human cells for it?

00:15:56: I think one of the big problems in animal experiments was using a model which isn't a human.

00:16:04: The promise with these NAMS or AI could be we just base our decision-making more on human data so our prediction about what would happen then as a real patient will become closer to reality.

00:16:17: That's something really exciting and that I think a lot of, not a lot but a couple of biotechs are already doing.

00:16:24: And having some big pharma partnerships on that using patient-specific xenografts so growing out single...using from different patients cells de-differentiating them then And that is, of course really interesting for capturing the heterogeneity between patients.

00:16:43: Because this again something we're lacking often in these animal studies and I mean if you imagine like these early studies were also not talking about hundreds or thousands of animals who are talking ten to fifty hundred mice which very close genetically mostly also having only one sex in there again to not make it more everything is very simplified and these models.

00:17:14: And of course, I mean we do that to get simpler answers but this also leads sometimes two problems later on because the answer was too simple and i think we need find a trade-off between being reductionist enough.

00:17:32: so biology is so complex that it's just very hard to make decisions based on it.

00:17:38: And this something we know, but at the same time also be complex enough.

00:17:50: topics or simple examples of that, we talked about a lot where again We have been two reductionists in clinical research involving humans.

00:17:59: is this huge gap of clinical Research when no females were used and trials actually also due to Thalidomide.

00:18:06: That was the trigger because you realize if there's embryotoxic and heterogenic compounds out There are chance that someone who might be pregnant would really bad.

00:18:17: So therefore, of course the reasons are always understandable but the outcomes that there is.

00:18:23: a number of approved drugs in clinical trials have never been tested on women which not what we want.

00:18:28: so capturing the heterogeneity with these numbs could just use cells from different individuals and different genetic backgrounds differences that can be captured.

00:18:44: And something that could be really interesting, and we talk a lot in this podcast about rare diseases right?

00:18:49: There you always think I mean there's has a genetic cause.

00:18:53: so all of the patients are the same because it is rare when your have a gene like okay but this isn't true even if they're in the same disease for example cystic fibrosis where It's the same gene involved.

00:19:08: You have so many different mutations that have very, very different phenotypes.

00:19:12: So even though it is rare and has the same cause The phenotype of these patients differs significantly.

00:19:20: Capturing by using patient-specific cells to test new compounds, for example gene therapies would be super interesting in that case.

00:19:28: And of course there is something where AI can come and integrate all this data that's generated with the numbs.

00:19:35: I think they're another thing.

00:19:38: these hybrid models as you say so linking in vitro and in silico together really very powerful because And we talk about that a lot.

00:19:50: The problem with AI is of course, that to be don't have the good training data would lack high quality trading data.

00:19:56: We have tons of data in biomedical research but it's unfortunately not off-the-quality that they need to train models and This promise that I just outlined.

00:20:07: so training things on human data or having better understanding of human biology Of course requires us to train it on human Data.

00:20:17: Experiments have, of course traditionally be done in animals.

00:20:21: So we need the human data to better capture that.

00:20:26: and here I think the NAMS can be a great bridging technology to even produce data... ...to train the models.

00:20:35: it becomes kind this virtual cycle where you at some point also could probably have this in a fully robotic fashion.

00:20:44: So we'll have optimized labs, something that a number of companies are already doing, recursion and turbine many others still investing in these kind of really autonomous labs where the robots carry out their experiments data is produced used to train AI they suggest new experiment somewhere.

00:21:03: there's little human in the loop.

00:21:05: itself doesn't make sense but having this cycle of data generation and training, the AI model can be extremely beneficial because you have high quality data.

00:21:19: You could answer very specific questions here.

00:21:22: we also had a chance to do something that I said in the beginning.

00:21:27: With the animal experiments, we often answer one question or maybe we'll answer ten questions.

00:21:32: But you should answer

00:21:34: a thousand for

00:21:36: may be a million questions where we can scale the number of endpoints and number of questions that we ask.

00:21:42: We can scale that really massively using these technologies And of course we can de-risk much more Using this and understanding better.

00:21:53: how would this compound really affect human biology.

00:21:58: What kind of toxicity would it potentially cause in different individuals?

00:22:02: Because what we also know, and this is something that we're often struggling to capture... And one of the main reasons why approved drugs have been withdrawn are toxicities.

00:22:14: Cariotoxicity but very commonly as liver toxicity.

00:22:17: The problem there's these rare sometimes idiosyncratic liver toxicities that you just don't capture in the animal studies because you have too few.

00:22:26: I mean, per se even you could capture it in the animals' studies but your only test at twenty or fifty rats and would've needed five hundred to capture it.

00:22:35: And then again with a clinical study You had only five-hundred patients.

00:22:42: Then if its drug thats used by many people might still get high enough rate of some genetic background, have some co-infection.

00:22:54: Have something in their biology that makes them or take another medication to react with this one and then have these rare liver events.

00:23:03: Capturing these things earlier would be really beneficial And again also having a more patient specific understanding.

00:23:10: does that actually work and is it actually safe?

00:23:13: So I think here we really have a lot of opportunities for using AI, and for using these new approach methodologies.

00:23:32: Two quick ones to wrap this up.

00:23:35: first off what your impression was?

00:23:37: the feedback both policymakers in industry you mentioned.

00:23:43: Also, of course the bold statement made by the FDA at the very beginning in that quote.

00:23:49: What do you expect?

00:23:50: Is this really going to take off right now

00:23:52: rapidly?".

00:23:54: So one thing I need say on the FDA's statement is that it was monoclonal antibodies not all drugs.

00:24:00: so a specific type of drug...I don't think its realistic for three or five years and we are really not there.

00:24:09: everything i said at the stage of huge potential and pilots that are done in biotechs, but they're for at least more sophisticated understanding.

00:24:24: so not just predicting with a simple model's toxicity.

00:24:30: Not validated enough to really decide if we can trust it.

00:24:33: And I think there is a long legacy or common understanding.

00:24:40: This is correct.

00:24:40: We trust this because the animal showed it

00:24:44: as to be seen in major papers, right?

00:24:46: Yeah and It's possible to replace that.

00:24:48: there are for example chemical irritation tests That used to be done in rabbits And was shown by us researchers that this can work better with AI or with computational models.

00:24:59: Now they're really completely reduced The use of rabbits For these tests and the models have been used.

00:25:04: so its definitely possible.

00:25:06: but I think the transition period will be really hard because we are still in a stage and there is for academic research people, as they're really believing also in papers worth more if there's animal research involved.

00:25:20: Because it's the model that shows something works.

00:25:24: then stepping away from that would take mindset shift.

00:25:29: but there was huge question of financial incentives.

00:25:36: facilities there are set up, they're running.

00:25:38: Nothing needs to be done.

00:25:40: changing this to new approach methodologies will require massive investments and we'll require training staff And well required having a common understanding of what that actually means in I think here the agencies I mean really also need To step it up and say What does success look like?

00:25:56: How can we qualify these type Of models In order to really Be on The same page about if We Can Trust them?

00:26:03: because at the current stage I would say this is really hard to say, can we trust these models?

00:26:09: Everyone who publishes this and has the develop set in their biotech will see it works.

00:26:15: And they'll work with hands but need something that's comparable between different approaches or different vendors for different labs.

00:26:22: just be sure that you can trust as much of animal model.

00:26:26: That said animals are not perfect at all.

00:26:32: these models currently are also not perfect.

00:26:34: So we're going to replace an imperfect model with another one, and we need to somehow make decisions where we see...

00:26:40: Making the new model a little more perfect?

00:26:43: Exactly!

00:26:44: And I think there's huge payoff – part of that is really the ethical pay-off because it gets great if you don't torture animals too much….

00:26:52: I'm not saying in a way that researches on purpose harm animals but every experiment put stress on the animal.

00:26:59: They don't obviously do not want to live under these conditions and everything that's being done for them will cause them distress, I mean this is very clear even under the most perfect condition.

00:27:09: Yeah i wanted to close on that note actually because of course we didn't extensively talk about the ethics aspect Because there are so many other things at stake And Of course There could be a comparison between animal testing and The meat industry For instance.

00:27:25: then We would produce Very different numbers For instance, I'm not saying this is a collateral damage.

00:27:31: But it is something that we are on the road of getting rid off right?

00:27:38: Yeah and This Is Always The Question Right That People Bring Up With How Can You Criticize Animal Experimentation If You're Not A Vegetarian Or Something?

00:27:46: I Think These Types Of What Aboutism Don't Really Lead You Anywhere.

00:27:50: I think This Is Not Weighing The Value Of Alive Animal or Human.

00:27:54: It's Always Very Tricky Ethical Debate.

00:27:58: I think the researchers luckily don't need to decide, but policy makers do.

00:28:04: And we have to give the researches frameworks in which they decide and it's great that this is really moving.

00:28:11: at the same time We also need be aware as you say of animal experiments being done for biomedical research are negligible In terms of numbers of procedures.

00:28:25: That said, I mean this is it's weighing right and i think every life counts in a sense.

00:28:33: And you need to just make your...I think as a consumer Make your decisions as a regulator make the decision and as a scientist also makes the decisions.

00:28:42: I think maybe on a personal level.

00:28:44: so for example For me it was really A decision to not do certain type of research because i didn't want To do that in my experiments, and they didn't Want to work with monkeys.

00:28:52: Because for Me personally this did not feel right?

00:28:55: I'm Not criticizing That other people are doing It because understand that They get valuable results out Of.

00:29:00: I could not do it.

00:29:01: So, i think that's really... It is a very tough debate and as many things we discussed there are a lot of grey zones where you need to be sensitive when discussing these things but if this can't be criticised then its too simple.

00:29:23: We really need to make smart decisions here.

00:29:26: And the cool thing about it is, It's not only a payoff on an ethical sense but in the end hopefully for industry be more efficient and make more money especially also for patients.

00:29:40: get better drugs.

00:29:41: That would be great!

00:29:45: Many, many thanks Izzy.

00:29:46: Thank you!

00:29:47: Any opportunities down the road?

00:29:48: this is a fascinating field and we have to readdress it in one fashion or another.

00:29:54: I think

00:29:54: Absolutely.

00:29:55: This was The Biorevolution podcast for this time.

00:29:58: hope you enjoyed the show.

00:30:00: further information as always

00:30:02: science-tales.com.

00:30:06: please feel free to add your comments.

00:30:09: encourage us to discuss on other topic.

00:30:12: thankyou again

00:30:13: Izzy.