AI is not going to replace humanity, or even chemists; like the printing press, however, it could be said that those who refuse to embrace it will be left behind by those who do, writes Caroline Maloney, a senior tech writer for the peripheral interventions research and development team at Boston Scientific Galway, and second-year student of science and technology at the University of Galway.  

The first industrial revolution mechanised production using water and steam power; the second harnessed electric power, we have had the third industrial revolution where we use electronics and IT to automate everything. Now a fourth industrial revolution is building on the third, the digital revolution that is now upon us.¹ 

Are people as scared today of artificial intelligence (AI) as they were when the printing press was invented in the 15th century? If I remember my history from secondary school, back then they destroyed the machines as they were considered harbingers of doom that would cause unemployment and the spread of misinformation.

Today AI has the same stones thrown at it. When we look at what happened after it took off however, the printed word did those things; but it also allowed for the spread of knowledge and created new industries. The point is that both scenarios are true, and are true also for AI – it is not a monster, it is a tool – as usual, whether it is used for good or evil will depend on the human beings wielding it. This article considers AI as a catalyst for good in our world as the fourth industrial revolution.

It is a machine intelligence that imitates how human intelligence works. The human brain exchanges data between millions of nerve cells and cognition happens through analysing data and identifying features. The algorithmic representation of this process is called Artificial Neural Network (ANNs or SNNs) and these are a subset of machine learning and are at the heart of deep learning algorithms. Their name and structure are inspired by the human brain, mimicking the way that biological neurons signal to one another.

So what is the difference between the cognitive power of a human to one of these AI programs? It is simply that humans find the recognisable features in the data accumulated from life experiences, but AI can learn from a vast amount of data beyond what is possible in human experiences. This is why many expect AI to revolutionise both our environment and our health.

Enter the ‘protein folding’ problem

There is (or was) an unsolved mystery at the core of scientific study that has been slowing progress for a half a century because it is hard to predict the shape of proteins, those miraculous little machines fundamental to life and chemistry. In July, 2021 however, two independent groups (the teams from Deepmind’s’ AlphFold 2 and RoseTTAFold) announced that they had solved it.

To explain it simply we can use our own make-up as an example; DNA boils down to a series of instructions for making proteins, and proteins are made from a set of 20 building block molecules called amino acids.

If proteins are like words, then we could say that amino acids are the alphabet. When the body makes proteins, it reads instructions from DNA to make long strings of amino acids which then fold in on themselves in particular ways and settle into specific shapes.¹⁰

That shape determines how the protein works because proteins need to fit together with other proteins; like puzzle pieces that latch onto specific molecules, and that makes proteins different from something like DNA, (where knowing the sequence is the same as knowing what it does) but for a protein the shape matters as well and the complicating factor for protein is that the way the amino acid string folds is determined by the sequence.

Different chemical structure

A protein can be made anywhere from 50 to 2,000 amino acids, and each amino acid has a slightly different chemical structure.

Adding to the complexity, individual parts of the amino acids can interact with all the other nearby amino acids and even some of the far away ones pushing the folding in seemingly random directions. So even though we know the amino acid sequences of billions of proteins, we are still stuck stumbling in the dark when it comes to working out their shapes.

X-ray crystallography is the traditional way of studying proteins, a process that could take days or even years depending on the protein. But in 2020, a team from the Google owned ‘DeepMind’ company showed that their new AI algorithm, AlphaFold 2, could predict the folded shape of proteins from amino acid sequences about as well as experimental methods and could do this is in a day or two.

Historically, predicting the folding of even a small protein took immense computing power, now AlphaFold, RoseTTAFold and their newer competitors can do this as a matter of rote. 

Speaking of competitors… while AlphaFold 2 and RoseTTAFold have similar accuracy, Meta's ESMFold is less accurate than AlphaFold, but significantly faster.

According to Meta, ESMFold does not achieve the accuracy of Deepmind's AlphaFold, but it is 60 times faster. This makes Meta's approach much easier to scale to large databases, as in the case of the now-published database of metagenomic DNA.² ESMFold inference is faster at enabling the exploration of structural spaces of metagenomic proteins.

Full complexity of interaction

The significance of this cannot be underestimated. The full complexity of interaction between cell-signalling molecules and gene-regulation mechanisms could now be reconstructed and elucidated for mixed species biofilms which would provide valuable data of ecological and medical importance³. 

A further example is the genomic analysis of 1g of soil. The coding capacity of the soil metagenome is greater than that of the human genome⁴. The potential this has for use in the environment is hugely exciting. 

Plastic is a great material; the problem how we deal with it at its end of life. If we could easily break it down by using enzymes – half of the problem at least would be solved (not creating, it in the first place is another task we could and should set for AI).

Enzymatic breakdown of plastics is low energy. It is environmentally friendly, and it allows us to recover those building blocks and infinitely use them over again. That means we can stop relying on fossil resources and reduce plastic pollution.^{5 9}

More sustainable enzyme approaches

"The possibilities are endless across industries to leverage this leading-edge recycling process,” says Hal Alper, professor in the McKetta Department of Chemical Engineering at UT Austin. “Beyond the obvious waste management industry, this also provides corporations from every sector the opportunity to take a lead in recycling their products. Through these more sustainable enzyme approaches, we can begin to envision a true circular plastics economy.”⁶ 

In 2016, Japanese scientists were sifting through the debris of a plastic bottle recycling plant. They were in search of a bacterium that could degrade plastic.

The type of plastic that the scientists were looking for was polyethylene terephthalate (PET).⁷ This type of plastic accounts for 12% of all global solid waste.

Estimates, say about 50 million tonnes are produced every year globally. By 2025, this could be a hundred million tonnes. As the demand for the material grows, so does the mountain of plastic waste in landfills. As PET based products are notoriously difficult to recycle. The goal of the Japanese scientists was to find a bacterium that could use PET as its energy source.

After bringing plastic samples back to a lab, what they found was a bacterium that amazingly eats plastic and converts it to carbon dioxide – it completely digested the plastic and used it to power its cells. What is left is carbon dioxide and the monomers, usually ethylene glycol and terephthalic acid – these components can make new plastics all over again.

Why haven't we used this everywhere? Well, there is one problem – the bacteria and even the enzymes weren't very practical; if the temperature was too low or too high, or the pH wasn't just right, it just wouldn't work.

Neural network

In May of 2022 a team from the University of Texas at Austin in the United States used a neural network to engineer and improve the performance and stability of the plastic-destroying enzyme. A machine learning algorithm modified the amino acids in the enzyme, the neural network called MutCompute studied a diverse range of 19,000 proteins of a similar size.

It was to learn the patterns of what makes a protein stable. According to Hal Alpa, who was the principal investigator of the study, amino acids that fit well within the protein are the key to stability.

An amino acid that isn't a good fit may be a source of instability. When this happens, the performance of the enzyme falls off. When the algorithm saw that an amino acid might not fit well, it suggested a different amino acid in its place for each of the 290 amino acids. The program checks to see if it fits well within its immediate structural environment compared to other proteins within its knowledge base. This means that it can balance the evolutionary trade-off between activity and stability.^{8 11}

Out of the millions and millions of combinations, the AI suggested three amino acid substitutions. The team decided to go with the AI's design. The result was highly, highly active, especially at lower temperatures compared to anything else that's out there. The AI designed enzyme broke down an entire plastic tray within 48 hours. It breaks down plastic at more than twice the speed and at lower temperatures compared to the next best engineered enzyme.⁹