- Binary Footprints
- Posts
- The Protein Folding Puzzle and The AI frontier in Biology
The Protein Folding Puzzle and The AI frontier in Biology
I was today years old when I found out that GPT-4 passed the United States Medical Licensing Exam with a score of 60%, which is nothing short of extraordinary. It also serves as an indicator that Artificial Intelligence is moving in the right direction towards actually changing the landscape of modern day medicine.
Let me start off by saying that I am by no means a biological expert and carrying this research out into the world of computational biology has caused me to develop a newfound respect for doctors and medical researchers. As a matter of fact, I did learn a lot while reading up on this topic which is one of the best parts of running this newsletter.
Like I said, AI really has started cementing it’s place in the world of medicine which should not be a surprise to anyone who isn’t living under a rock. My aim today is to show you why you really should be somewhat optimistic on what AI has to offer in medicine which is by far one of the most important aspects of life.
Google’s DeepMind really is at the forefront of this with what is called AlphaFold. To give a little context, I guess I could say this started with AlphaGo with is a deep learning AI that Google focused at an abstract strategy board game called Go. The aim of the game is simple: To capture more territory than the opponent by fencing off empty space. The problem is, this game, like I said, is deeply abstract which makes it the perfect place to train this AI. The AI was trained by viewing many old games on Go and after a while, it could examine between possible moves and determine which would lead closer to victory. Now, we’re presented with AlphaFold which is a more general purpose that can be applied not only in games but into real life problems like medicine and human biology. AlphaFold uses deep learning techniques to analyse a lot of data which helps determine the structure of proteins.
But why is this structure so important?
Proteins control a staggering amount of human functionality. From immune responses when fighting disease to even gene regulation, proteins play a very crucial part in human physiology. Like most people know, the functions of all these proteins are very closely linked to their structure so knowing their structure will be the first step in knowing what some of them do. However, a problem faced by many modern day biologists is protein folding. This involves folding these proteins into their 3D form which will give us an idea of their function. Folding these proteins, like I said, is no easy task and could take months by humans but with the introduction of computational systems like AlphaFold, this process is nowhere near as time consuming. The coverage level for all the human proteins is 98.5% by AlphaFold.
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/2b1e1a57-42ba-4505-9b6a-b5fdafedf23b/AlphaFoldImage.png?t=1717795060)
Three Dimensional Representation of a Protein
A benefit presented by this newfound speed is in drug production as AlphaFold can detect how drugs will interact with proteins in the body and also the efficacy these drugs offer. This could allow drugs to come out way quicker than before which will really do a lot of good for us.
So where does that leave human doctor’s?
Well, if it wasn’t obvious already, these tools are just that: tools. They can’t really do much without the doctors, scientists and pharmacists that implement them into their work. They serve as a means of helping and speeding up processes that would normally take way longer. But in the end they can’t function on their own which is why it’s important to see the things as tools and not replacements.
An example of this is Nuance’s Dragon Ambient eXperience (or DAX for short) which is a means by which doctors save buckets-loads of time on documentation. It involves DAX capturing the dialogue between physician and patient at any point in time which saves time and both sides and really allows the physician to pay more attention.
In the end, my ideal future is merging the best these AI tools have to offer like accuracy and access to much data with the experience and feel that human doctors bring to the table in order to improve life in the long run.
Also I apologise for the weeks without posting and as always, thanks for reading and see you next week.
Also if you’re interested in learning more about these topics as I don’t think I did that good at explaining in detail check out these links
Some news about AlphaFold implementation - https://www.isomorphiclabs.com/articles/rational-drug-design-with-alphafold-3#:~:text=The%20promise%20of%20AlphaFold%203%20for%20science%20and%20drug%20design&text=Already%2C%20we're%20using%20AlphaFold,effectively%20to%20a%20target%20protein.
GPT passing medical licensing exam - https://healthitanalytics.com/news/chatgpt-passes-us-medical-licensing-exam-without-clinician-input