Protein folding is a vital cellular process, that can dictate many physiological functions. A protein's three-dimensional structure holds biological importance as numerous diseases and syndromes can result from folding errors. 

The research community has been interested in identifying the 3D structures proteins for a long time as a proteins' shape is thought to dictate its function. By determining a protein's shape, its role within the cell can also be guessed, and scientists can develop drugs that work with the protein's unique shape. However, predicting 3D structures of proteins from their amino-acid sequences remained as one of the biggest challenges in biology,

Google-owned AI startup DeepMind's AlphaFold AI has now accomplished the ability to successfully predict protein structure, according to a recent paper published in 'Nature and PROTEINS'. Even though this isn't the first time, AI has been used to determine a protein's folding structure, AlphaFold's 3D model predictions are more accurate when compared to previous models. 

 "The 3D models of proteins that AlphaFold generates are far more accurate than any that have come before—marking significant progress on one of the core challenges in biology," states DeepMind in a blog post.

AlphaFold accomplishes this by predicting the distances between pairs of amino acids while the second technique predicted the angles between chemical bonds that connect those amino acids.

DeepMind trained separate neural networks to predict the distribution of distances between every pair of residues in a protein as well uses all distances in aggregate to estimate how close the proposed structure is to the right answer. AlphaFold then uses a scoring function to search the protein landscape to find structures that matched its predictions.

AlphaFold, debuted in last year, uses a combination of two existing techniques in the filed to "beat established contenders in a competition on the protein-structure prediction by a surprising margin."

"Our first method built on techniques commonly used in structural biology, and repeatedly replaced pieces of a protein structure with new protein fragments. We trained a generative neural network to invent new fragments, which were used to continually improve the score of the proposed protein structure," explains the blog post

The second method optimised scores through gradient descent, which was then applied to entire protein chains rather than to pieces that must be folded separately before being assembled into a larger structure, to simplify the prediction process.

 The code for AlphaFold is available on Github for learning more or replicating AlphaFold's results.