It’s hard to go a day in this modern life without Artificial Intelligence, or AI, sneaking in in some way, shape, or form. Whether it’s simply generating slightly creepy images for social media, simplifying the wording of scientific writing, powering internet search engines, affecting traffic on our website, or the unhelpful warnings about machines becoming sentient themselves and/or spelling the end of humanity.

But if we take the drama out of it, let’s focus on what it really means today.

The bAIsics

AI is a form of computational pattern matching. In order to make a decision, we need to collect data about that decision to inform it. This data is fed into the AI computer models. Those models seek to find mathematical patterns. If patterns are found, the AI system will then use these patterns when encountering new data. When the AI system is presented with new data, the patterns based on the “old” or historical data are applied to make a decision about it.

Much of the current concern about its use focuses on what data is used to feed the model, where it has come from, and how it is acquired.

ReplAIcing Animal Research

We recently had an internal meeting to discuss current and future use of AI that impacts on our work;

• How we currently use it

• What concerns we may have

• What our stance is/should be on its use in different areas of activity

It was a chewy and enlightening conversation, in which we could all raise our own experiences around what’s good, what’s bad, what’s useful, what’s ‘meh’, what’s overhyped, what’s scary, and what – ethically – we should do about it as an organisation, if anything.

Me, Myself and AI

I have the nagging sense that doing as much cognitive exercise as possible at my age is probably good for my neural plasticity, so I don’t often choose to use AI to help me, but I can’t avoid it completely in this line of work, and it does have its uses.

AI is used widely in science, research, testing and education, and our reliance on it will only increase, at least while we still have Earth’s resources to power it (but that’s another conversation entirely).

As it is now incorporated into internet browser search engines, I sometimes use it to find answers that I am looking for, but then need to fact-check everything that it throws up. Which takes time. And brain power. As well as brAIn power.

ReplAIcing animals in biomedical research?

I’m starting to annoy myself with these titles, but they’re better than those suggested by Google Gemini…I asked for funny AI subheadings, but they ended up being neither of those things. Algorithms, eh? No sense of humour. 

More seriously, AI is now used to speed up stages of drug development improving accuracy and efficiency in identifying potential drug activity and toxicity, and in some cases can be more effective than traditional in vitro screening assays.  AI-powered tools can quickly analyse vast datasets, predict toxic effects, and streamline the screening process, reducing the number of tests directly.

AI is not only simulating how cells might respond in vitro, but is also being used to simulate how animals might respond in experiments too. With clearly defined targets of the biological system being modelled and robust data, AI predictions can be appropriately translated into regulatory decisions, eventually taking animals out of the picture.

In the United States, steps are already being made towards this in regulatory decision-making for new therapies and drugs. Using monoclonal antibodies as a starting point, the Food and Drug Administration (FDA) is replacing animal testing in their development with more effective, human-relevant methods. The FDA’s animal testing requirement will be reduced, refined, or (hopefully) replaced using a range of approaches, including AI-based computational models of toxicity and cell lines and organoid toxicity testing.

Even in complex areas of research, like neurology, it is also predicted that brain organoids, computational models, and AI will revolutionise research and reduce reliance on animal models.

Searching for ReplAIcements

AI is also being used to help researchers search for ways to replace animals in their work. In the past couple of years, several AI-driven tools specifically aimed at finding 3Rs information and identifying alternative models have been developed.

1. 3Ranker is the result of a collaboration between the Karolinska Institute, SYRCLE and TenWise and was financed by the Swedish Fund for Research Without Animal Experiments. It searches for published information on all 3Rs across MEDLINE, the National Library of Medicine’s (NLM) premier bibliographic database that contains more than 31 million references to journal articles in life sciences, concentrating on biomedicine. 3Ranker is trained on a curated set of abstracts by human experts and will allow continued scoring of new abstracts as they are added. Users can search by keyword.

1. SMAFIRA is different. It aims to be a search engine for supporting the retrieval of alternative methods to animal experiments.  Users can search for alternatives based on an example in vivo publication that they give as an example. The algorithms predict the type of experimental model used in similar publications identified by scanning PubMed®, the NLM’s database of more than 38 million citations for biomedical literature from MEDLINE, life science journals, and online books.

1. This year, the EU Reference Laboratory for alternatives to animal testing (EURL ECVAM) launches their own BioMedical Models Hub (BimmoH), an automated database that collects and organises information on non-animal models used in biomedical research. BimmoH is designed to extract, filter, and categorise the scientific publications that use human-biology-based models, allowing the use of non-animal approaches.

The main goal is to create an accessible resource that promotes the use of human-relevant models and improves their clinical transferability. The database targets a broad user group — from researchers to project reviewers and funding agencies — providing easy access to model-related information across various disease areas.

The AI sorts the models according to three main categories:

• Anatomy, histology, and cell types (e.g., liver, brain, skin)

• Clinical conditions and disease areas (e.g., cancer, respiratory diseases)

• Model type (e.g., organoids, in silico models, organ-on-a-chip)

FAIvourable outcomes

We were lucky to have a sneak-peek test-drive of the Beta version BiMMoH ourselves last week, and to try and put it through its paces, I carried out a parallel search manually across PubMed, like the Old Skool puny human that I am.

This isn’t the place to review these tools, but suffice to say the technology – and its usefulness – is evolving rapidly.

We are all very hopeful that it will not only help speed up the process of finding suitable non-animal models, but also the uptake of their use across the research community, and of course, the collective movement away from the continued use of animals.

Whether you are a human animal, a non-human animal, or a robot, it is an exciting time to be in this line of work, and I for one welcome these AI-driven Animal Replacement Overlords.

What do you think of AI? Drop us a message:

Curious about how we’re using AI? Read our AI policy