Chemistry's AI revolution

BY NEIL EISBERG, EDITOR

As a tool for handling, investigating and analysing vast amounts of data, AI has no equal. The chemical industry, and particularly the pharmaceutical sector, has come to appreciate the impact of AI in terms of identifying new compounds, leading to new materials and therapies as well as new routes to manufacture for both new and existing compounds.

Projects using AI are to be found right across the board and it’s something we take a closer look at this month. In this issue (p22) we report on how researchers at the University of Pennsylvania, US, have used AI to trawl the ‘extinctome’ – the 208 extinct organisms with sequenced genomes available to science – to find promising peptides, including some from the woolly mammoth. We also look at how scientists are training AI to be better at chemistry (p26), and report on an AI-powered robot dog being used to hunt down the nests of invasive fire ants – and doing a better job than humans (p40).

In terms of chemistry and material science, AI is expected to transform the search for the next ground-breaking polymer. At Georgia Tech in the US, a professor in the School of Materials Science & Engineering, Rampi Ramprasad says: ‘In the early days of AI in materials science, research in this field was largely curiosity driven. Only in recent years have we begun to see tangible real-world success stories in AI-driven accelerated polymer discovery.’ 

But the old tech saying: ‘garbage in, garbage out’ is never more true than when dealing with AI predictions for new chemicals and processes. Their accuracy depends entirely on the availability of rich, diverse and extensive data sets – high quality data is essential. In addition, the design of suitable algorithms able to generate realistic polymers that can be synthesised is a complicated exercise. 

Despite these potential obstacles, the team at Georgia Tech has successfully used AI to produce designs for new materials to then be synthesised and tested to demonstrate whether they can be scaled up for commercial use.  

As fellow collaborator, Professor Ryan Lively from Georgia Tech’s School of Chemical and Biomolecular Engineering, notes: ‘We extensively use the machine learning models Rampi’s team has developed. These tools accelerate our work and allow us to rapidly explore new ideas.’ 

One success has been the design of new polymers for capacitors that store electrostatic energy. ‘The new class of polymers with high energy density and thermal stability is one of the most concrete examples of how AI can guide materials discovery,’ says Ramprasad.  

One criticism of AI is that it’s not always easy to understand how a model makes the decisions it does. And it turns out that taking a closer look at AI decision-making may be a source of scientific discoveries, too.

Researchers at the University of Manitoba, Canada, have used Explainable AI (XAI) to scrutinise predictive AI models more closely, presenting their results at the 2024 Fall Meeting of the American Chemical Society. 

Using an XAI model developed at Germany’s Karlsruhe Institute of Technology, they investigated the specific parts of drug molecules that resulted in the models’ prediction of compounds with biological effects.

‘Many chemists think of penicillin’s core as the critical site for antibiotic activity,’ said Rebecca Davis, a chemistry professor at the university. ‘But that is not what the XAI saw.’ Rather, she notes that it identifies structures attached to the core as the critical factor. 

‘AI causes a lot of distrust and uncertainty in people,’ says Davis. ‘But if we can ask AI to explain what it is doing, there’s a greater likelihood that this technology will be accepted.’ 

While AI is showing itself to be useful in the design of new materials and compounds, its applications have spread much further than pharmaceuticals and into general healthcare, which can bring in more issues.

For example, just what the public thinks about AI in healthcare has been investigated in a national US survey commissioned by The Ohio State University Wexner Medical Center.  

The centre is using conversational, ambient and generative AI to securely listen to doctor-patient interactions, with the patient’s permission, and to draft clinical notes for the patient’s electronic medical record for future review and editing. 

Generally, survey respondents were comfortable with the application of AI in healthcare. Of the survey sample, 75% believed that using AI to minimise human errors is important. When used in appointments with doctors, 70% were comfortable with AI taking notes and 71% would like AI to reduce wait times. 

However, overall, the survey showed that over half of respondents (56%) still found AI ‘a little scary’ and 70% have concerns about data privacy. And these views are echoed by the public around the world. 

Perhaps we should remember one thing: while AI is already proving useful across the scientific discovery process, there’s still a need to make sure its impact is explained and understood, as well as felt.