DeepMind’s Alpha-Fold Solves Protein Structures Using Neural Networks

DeepMind’s new improved AlphaFold AI has been selected at the biennial Critical Assessment of protein Structure Prediction (CASP) as the best at predicting protein structures.

Proteins are an important and fundamental part of the biological framework of the body.  it regulates almost every process of life. The proteins can be made up of large complex molecules or be simple molecules of less chain of amino acids.  Science so far knows that proteins fold it to get incorporated into the tissues and biological reactions. Protein structure prediction has been one of the grand challenges of biology.  However, with AlphaFold DeepMind seems to be one step close to real life protein prediction with the aid of deep learning neural networks.

protein alfa fold

Solving a 50 Year Old Problem

AlphaFold which first made its debut in 2018 initially relied on a brute force comparative strategy to predict proteins. However, inbuilt was a complex computational approach that used deep learning using vast data troves of sequences and structures of known proteins. This deep learning approach combined with “attention algorithm” helped AlphaFold to train its neural networks and and learn to spot patterns quickly.

The “attention algorithm” is developed to mimic the way a person might assemble a jigsaw puzzle: first connecting pieces in small clumps—in this case clusters of amino acids—and then searching for ways to join the clumps in a larger whole. Working with a computer network built around 128 machine learning processors, they trained the algorithm on all 170,000 or so known protein structures.

The results were phenomenal, across all the given challenges, target proteins by AlphaFold achieved a median GDT score of 87 to 92.4. The AlphaFold team was even given specific challenges to ensure its algorithm wasn’t cheating – it still excelled. Best part is that it was able to predict structures of proteins that are placed deep within cell membranes. These proteins are central to many human diseases and are difficult to visualize with x-ray crystallography.

Conclusion

One area where AlphaFold failed was predicting protein structures where the amino acids distort each others’ positions as they assemble. The AlphaFold team is next taking on this challenge.