Skip To Content
Cambridge University Science Magazine
A classical problem in structural biology is the ‘protein folding problem’ – that is, figuring out what final 3D conformation a protein takes. The 3D shape of a protein is important as it is closely linked to its function, but there are many theoretical ways that one protein could fold into its final shape. Biologists have traditionally resorted to expensive and laborious techniques for solving the problem, such as nuclear magnetic resonance and x-ray crystallography. However, DeepMind’s revolutionary artificial intelligence (AI), AlphaFold, could provide a much more accessible solution. 

British AI company DeepMind, co-founded by Cambridge alumnus Demis Hassabis, specialises in designing neural network systems: each is a series of computer algorithms that collectively act like a human brain, in that they can extract and learn underlying relationships in a dataset and apply this to new data, albeit much more powerfully. Learning from past examples, AlphaFold is able to take the sequence of amino acids that make a protein – its building blocks – and predict the final 3D structure with incredible accuracy: better than all other teams that entered the CASP14 protein folding contest and very close to the experimentally determined structures. 

Although experimental data still remains the gold standard for determining protein structure, AlphaFold’s immediate impact will be to reduce the amount of data needed to reliably predict a protein’s 3D shape, empowering research that was simply not feasible beforehand.

Adiyant Lamba is a second year PhD student studying developmental biology, and News Editor for BlueSci.