.png)
By Lakshanya P.
How can we predict the 3-D shape of a protein? In the science world, this is known as the protein-folding problem. Christian Anfinsen hypothesized that the 3-D structure of a protein would be determined by its primary amino acid chain and the environmental conditions during the folding process. He shared the 1972 Nobel Prize in Chemistry for his hypothesis, known as the Thermodynamic Hypothesis, which created the Anfinsen dogma. However, solving the protein-folding problem isn't easy, as it involves a massive computational challenge known as Levinthal's paradox. With countless possible ways proteins could fold based on their amino acid sequence, finding the correct structure used to seem impossible. Scientists had no choice but to spend hundreds of thousands of dollars, and work really hard, just to study a tiny fraction of the 200 million different proteins we know of now. The computational approach that the scientists first used to solve this problem was the Monte Carlo simulation, and this was actually a rather effective way of tackling this problem. However, Google scientists found an even better approach. They came up with alpha fold, to predict protein structures accurately and rapidly and this is where the AI comes in. The release of alpha fold took a long time and took many different tries to find the best approach—let’s go through a quick timeline. In 2016, Google started working on AlphaFold.
In 2018, DeepMind used deep learning techniques to predict protein structures more accurately than before. Deep learning is a branch of AI that mimics how our brains learn. AlphaFold leverages deep learning algorithms, like attention mechanisms, so that it prioritizes the interactions between the amino acids that are crucial for accurate folding predictions. The next iteration, called AlphaFold 2 and released in 2020, 2 years later, was a game-changer in predicting protein structures. There was a huge jump in accuracy and it was almost as precise as a person running an experiment. Then, there were even more updates in 2024. AlphaFold 3 is the iteration that we currently use right now. Here’s why it’s so exciting for us! First, the accuracy was improved even more. Next, alphafold 3 can predict structures of protein complexes that are vital for understanding biological processes involving multiple interacting proteins. It also considers PostTranslational Modifications like phosphorylation. These modifications can totally change how proteins function, so predicting them accurately is a big deal. On top of that, It's not just about static structures anymore. AlphaFold 3 predicts how proteins move and change shape in different situations. This is crucial for understanding how proteins behave inside our cells.
Commentaires