We are currently in the middle of some of the most impressive healthcare accomplishments in the history of mankind. By now I’m sure everyone is acutely aware of how quickly a COVID-19 vaccine has been produced as compared to historical norms. I’m sure you’re also aware of how scientists and researchers used new technology – called messenger RNA to enable these breakthroughs. I’m not here to talk about that or opine on the value of the vaccine in general. Instead, I wanted to marvel as another incredible breakthrough that we’ve seen in the last couple of weeks and something that hits a lot closer to home for me.
I’m incredibly bullish on the changes and opportunities that Artificial Intelligence (AI) and Machine Learning (ML) create for our society. When I see or read about a paradigm-changing innovation, it reinforces my love for the science and technology behind what we do. While I realize our ML models don’t save lives, I still think it’s fair to look with pride as other AI researchers accomplish tasks that even five years ago would have been unthinkable.
We had one of those this month when Alphabet (the company that owns Google) announced AlphaFold – an outgrowth of their DeepMind AI company– has the potential to completely change the rate of acceleration of research in every field of biology. At the risk of being too technical, I think it’s incredibly interesting to take a quick look at what they were able to do:
Every living being is made up of tiny molecular proteins that guide every single biological process our bodies perform. They are the literal building blocks of life. There are currently about 200 million of these proteins, but as this is a relatively new area, we are able to study, about 30 million are discovered every year. Each with a unique 3D shape that determines what it does and how it works. However, as you can imagine, understanding the structure of these proteins is very time-consuming and expensive, so we really only understand a very small fraction of the proteins we know about – fewer than 200,000 out of the 200 million. Gaining more knowledge is key to understanding disease, medical treatments, and the mysteries of life as a whole.
If you unravel one of these tiny proteins, what you’d see is basically a string of beads made up of a sequence of chemicals known as amino acids. The specific amino acids in one of these strings causes the protein to fold into a unique shape and structure. Scientists have spent decades trying to determine the exact structure of a protein based on the sequence of amino acids. This challenge is known as the protein folding problem.
What Alphabet did to try and tackle this problem was train a machine learning model on the sequence and structure of about 100,000 known proteins. Based on the training in this model, AlphaFold is now able to make incredibly accurate predictions of what shape a protein will form based just on the sequence of amino acids. Work that would have taken months or years before can now be done instantly.
I started this post talking about COVID-19, so let’s look at that to get an idea the impact of this research. SARS-CoV-2, the virus that causes COVID-19 is made up of about 30 kinds of proteins – ten of these were very poorly understood. This makes it hard to prevent, treat and cure of course. The team at AlphaFold accurately predicted the structure of six of these proteins back in March. For comparison, the structure was finally worked out through traditional means just a few weeks ago. This approach to medical research is going to enable us to not only respond rapidly to new and emerging threats and diseases, but also circle back to study problems that have been previously underserved.
To put a little finer point on it, it can take in excess of $2.5 billion and 10+ years to develop a new drug. The research from AlphaFold could cut that cost and time by a factor of more than 100, leading to previously inconceivable speed and efficacy. We are truly on the precipice of a new age of biological breakthrough. And it was all made possible through a machine learning model with only 100,000 examples. That is simply amazing!
I hope to spotlight other incredible innovations powered by AI and ML in the weeks and months to come. I love this area of study and I hope you’ll follow along as we dive deeper into it. Also, we’re going to create our own version of mythbusters where we look into technology or solutions that claim to be AI but are really just a bunch of if… then statements. That should be a lot of fun too, so please subscribe and follow for lots of great content!