| Article by Sophia N. |

While the public marvels at chatbots and digital art, AI has been quietly cracking one of biology’s hardest problems: protein structure. Long invisible and stubbornly complex, these molecular forms govern life itself. The future of medicine, it seems, may owe more to code than to laboratory bench.

Over the past seven decades, the tedious task of determining the complex protein structures found all around us was manned by hundreds of thousands of biologists, producing around 150,000 structures manually. The turning point came in December 2020, where a team of 15 biologists managed to verify over 200 million proteins and won first place in the CASP competition. The method is simple, small, and a concept that invades most 21st century conversations: artificial intelligence.

The Critical Assessment of Protein Structure Prediction (CASP) is a worldwide community experiment held every two years since 1994. It provides researchers with an unbiased, competitive opportunity to assess their structure prediction technologies, with the top models published for scientific use. 

Operation is simple. With recently discovered “target proteins” being kept secret from participants and organizers, submitted predictions are weighed against said targets using a special criteria. The models are then scored from 1 to 100, with 1 being the least accurate and 100 representing a perfect match to the target protein. In the first year, the highest score was 40; from the 5th to the 12th competition, scores plateaued between 70 and 80. In the most recent CASP14 competition, however, scores peaked at a whopping 92.4.

This incredible feat was achieved using the AI program AlphaFold, developed in 2018, and its successor, AlphaFold 2, released a few years later. Both programs utilised deep machine-learning and neural network technology trained on public protein databases. These two concepts allowed the AlphaFold program not only to learn the key structural identifiers of proteins, but to build completely new, unique and scarily accurate predicted protein structures.

Naturally a certain question occurs: why focus so heavily on proteins? Put simply, proteins are the fundamental building blocks of life. Constructed from sequences of just over 20 distinct amino acid monomers, these chains can fold into billions — if not trillions — of unique three-dimensional structures, each tailored to a specific biological task. A key class of these proteins are enzymes — biological catalysts that dramatically accelerate vital chemical reactions, such as digesting nutrients or speeding up microplastic decomposition. Yet, catalysis is only a fraction of their impact.

It begins to become clear how invisible structures hidden in our everyday lives can be manipulated not just to aid humanity, but to advance it. Enzyme based anti-bacterial medicines will be stronger, plastics will decompose faster, and toxic waste can become a thing of the past.

It seems ironic that the same framework a lazy high schooler might use to write an essay can solve some of modern humanity’s biggest problems. However, innovations such as AlphaFold highlight the importance of integrating new inventions into key fields, setting a precedent for a future where we do not live in fear of AI, but rather work and develop alongside it.