AI threatens yet more jobs – now, lab rats: Animal testing could be on the way out, thanks to machine learning

Machine learning algorithms can help scientists predict chemical toxicity to a similar degree of accuracy as animal testing, according to a paper published this week in Toxicological Sciences. A whopping €3bn (over $3.5bn) is spent every year to study how the negative impacts of chemicals on animals like rats, rabbits or …

[Adrian 4]

Sounds completely useless.

Isn't the point of testing to find out if there are effects that aren't expected from the theory ? If the computer models were perfect, there wouldn't be a need for testing at all. So relying on a knowledge base - even one with trendy words attached - isn't going to do even half a job.

[Anonymous Coward]

Yeah, but shinny 'AI' thing here...

You know the media loves me right now....

[a_yank_lurker]

@Adrian 4 - I see even more problems with this. It is well known that trans species toxicity varies widely for a specific chemical or class. So what is nasty for your average lab rat is not necessarily a problem for a human. Also, there have been accidental discoveries that occurred when developing a new pesticide or drug. The original nerve agents were developed as pesticides and were found out later to be rather toxic to mammals including humans; a point you made. Viagra was developed as heart medication and during clinical trials it was found to be much more effective for erictile disfunction; reinforcing your point.

Most people do not understand to role of animal testing, which is probably less than believed. It is understand the underlying basic biology and chemistry on a mammal. On occasion these tests will require doing things that can not be done ethically on a person thus the need for some animal testing. (Not to deny that some animal tests seem to be ethically challenged at best, another discussion for another day.) But having a clue how something works in a rat gives a clue of how it will work in other mammals.

Also, I suspect these Artificial Idiocy systems are not actually modelling the various reactions (which may not be known) but whatever toxicological data is available. Thus, the masses will think it will be a perfect substitute for animal testing when at best it will give an indication of what to look for.

[Mark 85]

Let's add one more to the list... aspirin. It can be deadly to many types of test animals.

[0laf]

I graduated in pharmacology, we were told by our prof back in the day that aspirin would never be approved if it was a new drug discovered today.

I assume since 20yr has passed since that comment it would be even more unlikely a drug that we think of as innocuous would be approved for human use now.

[Anonymous Coward]

>I graduated in pharmacology, we were told by our prof back in the day that aspirin would never be approved if it was a new drug discovered today.

That is true and also a huge problem. Any new drug that kills all rats or dogs etc. will see the research project terminated pretty quickly. Normally. Yet we know that things like onions are healthy for humans and deadly to dogs. Same with chocolate and a dozen other common foodstuff. And there is no easy way around this while many realise the lab books could be full of terminated projects that could have been beneficial to humans.

Certain companies with flexible moral standards have seen a way around this: testing in slums. A front is set up, people are paid and results recorded. If things go south the front is burned and no connections can be found between the company and the illegal tests. If on the other hand a success is found the company will have to falsify years of lab animal testing since they already know it works on humans but kills animals. And again the front has to be burned to avoid the whole thing being uncovered.

The whole patent cliff issue appears to make more companies willing to break the law in order to find the next great blockbuster drug. Only a few have been caught and tried but there has been remarkably little about it outside specialised news channels.

[Allan George Dyer]

@a_yank_lurker: "effective for erictile disfunction; reinforcing your point"

Yes, yes it does.

[Nick Kew]

@Adrian 4

When the question is effects on humans, how is testing on lab animals any less useless?

This is an argument that's been going on for years. Bottom line is, animal testing is expensive, and therefore commands high research budgets that sustain shiny careers.

[spold]

>>>Sounds completely useless

No, AI Robots will be able to predict rust... very useful for them

[Allan George Dyer]

Only 57%?

"experiments resulted in the death of the poor critters 57 per cent of the time"

You have to use fresh animals for each test (so previous exposure doesn't influence the current test), and I don't see retirement homes for ex-lab rats, or a big demand for ex-lab rats as pets. So, don't the survivors get killed at the end of the test?

[Schultz]

Re: Only 57%?

The toxicity of chemicals in the Chemical Safety Data Sheet (CSDS) is given as LD50 value (lethal dose causing 50% mortality). I assume the additional 7% die from natural causes.

The big question I didn't see addressed in the article is whether the 'predictive power' was evaluated based on chemicals that were part of the training data (making it more of a recognition AI) or whether the predictions were made for compounds that were not part of the training data.

[Allan George Dyer]

Re: Only 57%?

The LD50 value is the experimental result, the percentage that die in the experiment depends on how you search for the value. For example, test 1, dose 10mg, result: 100% survival; test 2, dose 20mg, result: 90% survival; test 3, dose 40mg result: 30% survival, test 4, dose 30mg, result: 50% survival. Over the 4 tests, the experiment has killed 32.5% of the animals. Another experiment on a different substance could have a much higher lethality. 57% is an average over many experiments, so maybe the experimenters are, on average, fairly good at predicting the toxicity of a substance.

But, you miss my point. Part of the procedure is, something like, 'at the end of the experiment, dispose of the consumables according to standard safety procedures and clean the lab', where 'consumables' includes lab rats, and the standard safety procedure is kill and incinerate. None of the animals leave the lab alive, even if they survived the dose.

On your big question, I would hope that the reviewers at Toxicological Sciences would reject a paper that only 'predicted' the training data, but it would be nice to see confirmation.

[I ain't Spartacus]

Re: Only 57%?

Our company did some work for an animal testing lab - and I'd argue they're not terribly logical people. The experiments which used cows were performed on the first floor, requiring a rather large lift - whereas I'd put the rats up there, and just use the stairs.

However, once you've experimented on your cows, presumably lab cleanup would be 1. Remove cows from lab 2. Get shovelling 3. Light barbeque 4. Get beer...

[Korev]

Re: Only 57%?

Our company did some work for an animal testing lab - and I'd argue they're not terribly logical people. The experiments which used cows were performed on the first floor, requiring a rather large lift - whereas I'd put the rats up there, and just use the stairs.

You're assuming the scientists had choice in the matter. We've all heard of datacentres being built in the basement of buildings next to a river, this is the same.

[sml156]

What's Cheaper

Testing using AI computer

Using lab animals

I'm guessing animals is cheaper so the majority of labs will continue using animals.

[I ain't Spartacus]

Re: What's Cheaper

Computers would be cheaper. Much cheaper. If you're testing on animals you have to buy them, then buy their cages and food, keep them healthy (so as not to bugger up your data) and fill out loads of paperwork to prove you're using them ethically.

the reason we don't use computers instead of animal testing is the same reason that we don't go straight from testing drugs on tissue samples in test tubes. We don't fully understand all the interactions of all the processes we're studying, and we keep finding that reality differs slightly from out models. Therefore we have to test on animals to learn what we don't know - even though some substances have different effects on different animals.

That's why after testing on animals, we usually test on healthy humans in tiny doses, so that we can then learn yet again where our models don't fit reality.

This is why it now costs tens to hundreds of milllions of dollars to get a new drug approved for sale. It's a very long, multi-stage process, and drugs are failing at animal or initial human trials all the time - either because it turns out they don't work as well as theorised, or because they're too dangerous.

We simply don't have the knowledge to automate this process, until we understand the underlying science better.

[I ain't Spartacus]

A small digression

If you will permit me to reply to myself with a small digression to waste your time...

I was watching a discussion program on the tellybox ten years ago. About animal testing. And there was a young animal rights campaigner getting very animated about how all animal experimentation was cruel and should be immediately banned and replaced with computer modelling.

Which is basically a bollocks argument, becuase of the points I made above. We don't understand the underlying biology well enough to do this yet. We're still making mistakes, even though we're getting better.

Anyway his argument got shot down by another guest who had severe Parkinsons disease. And who said that his treatment had been developed on live gorillas - basically they practised the brain surgery techniques on them first. So the most problematic type of animal research - vivisection on higher primates.

While he was saying this he reached into his pocket and pressed a button. And he instantly transformed from a normal bloke, talking and gesturing, into this hunched and totally rigid figure, barely able to control his movements. Then he very slowly inched his hand back to his pocked and pressed the button, and instantly transformed back. He's got electrodes wired into his brain, which he'd turned off - which I hadn't even realised was possible at the time. And they stopped his muscles from going into involuntary spasm.

Our understanding of the brain on a physiological level is still pretty low sadly, so I can't see us being able to replace animal testing for drugs or surgical procedures for decades to come.

[Dagg]

C# or Java programmers

Hey, my 5 pet rats would make perfect C# or Java programmers.

[I ain't Spartacus]

Re: C# or Java programmers

Surely rats should program in Python?

Or at least my pet python thinks that would work nicely for him...

[lazylizard]

is it new?

http://web.bii.a-star.edu.sg/~loolh/

[Chris G]

Yet another piece of chairbound research digging up statistics and trying to call it science.

To be really meaningful it would have to be run along side many different studies using animals for years, to both teach it and compare and develop it. Even then I doubt it will equal the complexity of effects that can be found in mammalian biology.

'Ardly Interesting!

[the Jim bloke]

web cartoon I cant remember the address of

Engineer: "Perhaps we should consider shutting down our animal testing program"

Manager: "Why ? Cosmetics companies use it all the time"

Engineer: "Yes, but we make hammers.."

[GIRZiM]

There's are reasons

why canaries are still taken down mines alongside all the whizzy, new tech that detects gases.

And it's not just because the batteries never run out on the canaries.

[PeterKr]

Lots of misconceptions here

Animal testing is cheap - Not really. For most tests animals are kept in highly controlled sterile environments to minimize variation. Expect a very short term study with a dozen mice to cost 5 figures, while a year+ will go for 7.

Animal testing is useless - kinda. Animal testing gives a lot of wrong answers about efficacy/safety in humans. Because we have been doing it for so long, we know a few ways to predict when the results might be suspect, but it is still a long way from perfect. Unfortunately, you can't just start out by testing in humans, and animal testing is the best we currently have.

[EveryTime]

So my post-retirement job (Silicon Valley, so when I can no longer pass as being in my 30s) first pick will be gone. I guess Walmart greeter it is.

[Anonymous Coward]

Simulation is not a new issue

There was a time when it was thought that one could simulate the effect of molecules to see what effects it had in order to develop new medicine. Years and years were spent on this and had it worked we would now have had a LOT of medical breakthroughs. Instead it failed. And scientists returned to looking in nature for new compounds and testing these on mice, rats, etc. all the way up to humans.

So I am surprised people were surprised by this result. It is old news.

[tonybarry]

There are some things even a rat won't do

Java ? There are some things even rats would prefer not to do.

Mine is the coat with the cheese in the pockets. Hmmm.

[Dagg]

Re: There are some things even a rat won't do

Mine are hipster rats, they will do anything for smashed avo even program in java.