AlphaFold: Protein Predictions Like No Other

If there's one machine learning technology that encompasses the brilliance of machine learning, the very reason it is on its way to becoming a pervasive technology, and also the most prevalent and important application in the entire world - health - it's AlphaFold.

Proteins matter. They don't mass. (Please laugh).

All building and reparation of cells, tissues, and organs in your body is catalyzed by proteins. Amino acids, the molecular machines of life, have one key characteristic that revolutinizes their function: folding.

The folding of linear amino acid chains is a physical process in which linear monomers of amino acids spontaneously fold from a miscellaneous array to a structured coil. The shape that amino acids fold into is inherently spontaneous, but it would revolutionize science to have predictive mechanisms to hypothesize what kind of shape the protein will make in its tertiary phase. Shape matters in biology and biotechnology; active sites in enzymes, receptor proteins, antibodies, and DNA rely heavily on the shape of the molecules around them to complete their function.

Scientists realized the value of predictive protein technology early, and to accomplish their goals, scientists relied on laborious techniques like X-ray crystallography and cry-electron micrscopy, and despite decades of efforts, only a fraction of known proteins has their structures predicted and resolved.

But then came Alpha Fold. https://deepmind.google/technologies/alphafold/

AlphaFold was training on a vast number of protein structures, identifying patterns of functional groups in amino acids that correlate with specific shapes. AlphaFold enables structure-based drug design, helping researchers target various diseases like COVID-19. During the pandemic, it provided insights into proteins of the virus that were critical to develop its vaccine. AlphaFold could accelerate enzyme design (you'd be suprised to known how much of the biological world runs on enzymes) and elucidate unknown protein functions. Potentially, if turned into a generative model, it could even create new proteins previously unimagined to scientists.

In my next post, I'll be talking about how AlphaFold and other machine learning networks can create and generate novel output. I hope you learned something cool about AlphaFold today, and I'll see you in my next blog post.