First, a recap on protein science! Proteins are made up of amino acid, a set of 20 distinct building blocks strung together like beads on a necklace. On its own, this string of amino acid doesn’t have much use, but become functional as this necklace of beads tangle up and form unique 3D shapes. Proteins get their shape from folding this linear string of amino acids into a 3D spatial arrangement. This would be known as the tertiary structure of a proteins, comprised of spirals of α-helixes and stacks of β-pleated sheets.
The structure of a protein is mostly determined by its amino acid sequences but looking at the linear string of protein building blocks isn’t helpful (at least to the human eye) in inferring the protein’s structure in 3D, akin to looking at a crochet pattern and visualising the shape of the final product.
Traditionally, protein 3D structures are determined experimentally by methods such as nuclear magnetic resonance (NMR), X-ray crystallography and cryo-electron microscopy (Cryo EM). Shooting waves like X-ray or beams of electrons at a protein and see what flies past and what bounces off. These methods may not be idea when inadequate funding and expertise is available. They cost a lot to perform and require highly specialised training. On top of them, each of these experimental methods have their own drawbacks. For example, there is a size limit to the protein if you call for an NMR, while X-ray crystallography does not work if your protein of interest does not crystalise. Collectively, it is also almost impossible with any of these methods to image the structure of protein that sit in the membrane as the purification process of these proteins may alter their configuration.
Here comes computational methods to the rescue, like the recent Alphafold developed by Google’s AI offshoot, DeepMind. Alphafold is a protein folding prediction software. It predicts the 3D structure by analysing the sequence of amino acids in query and compare it to the folding patterns of proteins with known amino acid sequence and 3D structure in its database. Alphafold is way above it’s competitors in predicting protein structures during the annual Critical Assessment of protein Structure Prediction (CASP) competition. In CASP 2020 DeepMind achieved prediction power significantly above with Alphafold and maintained a rather steady performance even in the more challenging stages of the competition.
Look, I know what you all are thinking: another workforce replacement by AI and automation? But we need not worry, Alphafold is a prediction programme and we still need to experimentally validate the protein structure. Fellow scientist, your job is here to stay. Alphafold isn’t here to take your job, they are here to be a pal in your research journey. For now, lets celebrate to arrival of a new comrade in structural biology.
Cheers. OwO
