How AI Solved One Of Humanity's Hardest Problems

this AI model has solved a problem thought to be impossible or at least decades away from being solved proving countless scientists wrong and the results could mean revolutionary changes in cancer treatment or even possibly in climate change it has to do with the way that proteins work which unless you're in biology or medicine is probably a foreign topic the word gets thrown around in all sorts of context are you guys talking about protein I love protein but that doesn't matter all you need to know about proteins is that they are incredibly important to virtually every process that your body carries out yet just like many things in biology proteins are a little bit of an unexplored territory that is until now thanks to AI we've now unlocked the ability to better understand and predict the patterns of proteins meaning endless new possibilities in the world of Biotech over the past couple of months I've spent endless hours researching the world of proteins all in hopes to show you how this project could better the lives of you me and the entire world so to talk about how we could develop cancer fighting proteins we have to start with a little bit of context you see proteins are actually insane the amount of shapes that they can form themselves into is astronomical to put things into frame people have been trying to better understand these protein configurations for more than 50 years somewhat unsuccessfully see they're made of amino acids and there are 20 amino acids that you can chain together in any order so as you can imagine the possibilities expand quickly let's say we have a chain of three amino acids that's already 8,000 possible combinations but we're not done there you can then take multiple chains of amino acids and bond them together making an entirely different type of protein and that protein can then fold itself into different forms in order to carry out specific tasks for example here is an educational video of someone building the structure of insulin insulin is the protein involved in regulating our body's energy Supply and blood sugar levels and its composition of amino acids is structured to do exactly that its two chains are folded together in a specific structure that allows it to properly function if the chains were oriented in any other way it wouldn't fit into its receptor and could not trigger the process of maintaining our blood sugar levels this folding is a key insight into understanding diseases like diabetes and cancer because that's how all proteins are they're folded perfectly for the task at hand a key for a lock but they're not just any order AR key imagine if you had a magic key that could adjust its grooves to match any lock millions of doors are now unlockable because of this one key that's a pretty powerful key and that's why proteins are pretty damn powerful as well only with a protein we're not talking millions of key holes that the protein can change shape for we're talking many many more let's take the entire timeline of the universe and place it on one line that's 13 billion years right there and now let's take a protein and have it fold into the configurations available to it let it shift into shapes constantly every day over the age of the entire universe now even if you attempted millions of configurations per day you still wouldn't be done with the possible combinations of that one Protein that's because any given protein can theoretically adopt somewhere around 10 to the 300 different configurations as we just demonstrated that's an insane amount of structures yet somehow each protein spontaneously folds into one specific shape and carries out its task accordingly out of all those combinations it knows what shape to pick how crazy is that but why do we care about protein folding how does this relate to cancer and diseases well let me show you this is axin one there's something different about this axin one though it's folded improperly it's supposed to look like this and the gradual increase in misfolded axin one actually leads to Alzheimer's if we could better understand the folding of this protein as well as what causes it to misfold we could better understand Alzheimer's and possibly come up with a cure but remember ax in one has enough possible structures to last you more than 13 billion years of trial and error it's no wonder that this puzzle was thought to be many decades away from a solution it's an astronomical task to predict protein shape and yet AI did it we have been stuck on this one problem how do proteins fold up if we can reliably predict protein structures using AI that could change the way we understand the natural world the Deep Mind machine learning program called Alpha fold Alpha fold Alpha fold meet Alpha fold an AI project created by Deep Mind it's an artificial intelligence company working under Google and in 2020 Deep Mind entered into this competition Casp it's an event held every year to see how well teams around the world can predict protein shapes teams are given a set of amino acid sequences and they have to do their best to determine what the corresponding protein might look like in 3D Space by the time the competition Begins the judges have already mapped out the answers but they did so using traditional painstakingly slow techniques the teams must come up with software solutions that can do the same but at a quicker pace and whoever can match their predictions to the actual model most closely wins first place now before Alpha fold a winning score was somewhere around 40% accuracy but in 2020 Alpha fold blew everyone out of the water with an accuracy score of 92% some structures were so close that the margin of error was the size of an atom they correctly guessed certain amino acid sequences so perfectly that out of thousands of atoms only a single atom was slightly out of place in other structures where the Deep mine team was farther off the judges noted that they weren't sure if that was because Alpha fold's model was off or if their estimate of the protein structure was off for the first time the judges didn't just award a team first place they also declared the protein folding problem solved but how did Deep Mind do it by the way if you don't care about how alpha fold works and you'd rather skip to the part where we talk about the amazing things that we'll be able to do with it you can go to the timestamp on the screen okay so what was so different about Alpha fold that allowed it to create a gap like this between them and the competition well it starts with Evolution you see we've mapped out a massive collection of protein structures within us and other species 170,000 different protein structures by assessing this collection and paying attention to other species in the evolution of their bodily functions we can get hints to where amino acids might end up in a structure if an amino acid is in a particular spot across species there's likely a reason for it to be there for example if an amino acid is hydrophilic meaning it loves water they are shown to be consistently on the outside of a protein since proteins are often surrounded by high water but if it's hydrophobic it's more likely to be in the middle of the protein now what if we zoomed out our scope just one more index what if we looked at amino acid pairs let's say you have two chains of amino acids you can line one chain up at the top of a grid and one to the left and then for each pair you assess the massive collection of existing proteins again but this time you look at every time these two amino acids are together so based off of that you now have another chart of potential positional data for the protein structure okay so now you've got data on previous protein structures doesn't every team have that well yeah so this is where AI comes into play machine learning is really good at taking an existing collection of data and then guessing how that data might be used to predict the creation of something new just think of chat gbt it's been taught all of the written texts found on the internet now it does a pretty good job at predicting the sentences bullet points and essays that you ask of it the same thing can be done with protein structures you can teach a model to somewhat understand protein structures and then ask it to try and build new proteins from chains of amino acids that it's never seen before and as you give it feedback it marginally improves over time now you've got a system that can more accurately construct proteins better than any human has in the past but now where does that leave us sure this all sounds great on paper Alpha fold solved a ridiculously hard problem but how could it actually yield benefits to you me and Society well in absolutely incredible ways mindboggling ways one of the most immediately obvious ways that this could be revolutionary is in designing synthetic proteins let's say this box represents all of the natural proteins known to us well then this grid represents all of the possible proteins that we could theoretically construct the natural proteins that do exist have been structured a particular way over millions of years and are really good at one particular task but nowadays we're living longer which has revealed new diseases diseases that no natural protein is properly designed to deal with why don't we build synthetic proteins today that can help us better human lives in the near future for example let's talk about this equation what do you think that this might mean we're talking about proteins a lot so you probably worked out that these are proteins but what's exactly going on well this right here is a protein from the respiratory virus RSV and this is what scientists call a host receptor protein when you combine the two you get a vaccine that creates a much stronger vaccine response than any of the other vaccines that have been tested AKA you get a better vaccine RSV is one of the leading causes of infant mortality worldwide this protein alone could save millions of lives but now look at this protein because this could be even more revolutionary look similar right only this protein is made up of several virus proteins and in this case these are flu viruses from around the world meaning you could create a universal vaccine a vaccine that could Target multiple strands of a virus at once it could be administered once and never need it again for an entire lifetime but this technology could be used for much more serious diagnoses as well like cancer one of the biggest issues with cancer is the fact that you can't Target cancer cells specifically to better explain this let's take a trip to the bone marrow this is what healthy bone marrow looks like these are baby blood cells growing and dividing these cells go on to carry oxygen from your lungs to the rest of your body but with chemo the bone marrow begins to look more and more like this and the Damage doesn't stop at bone marrow this widespread form of treatment can cause permanent damage to the heart lungs nerves and kidneys but with custom-designed proteins we could deliver chemotherapy directly to the cancer tumor avoiding the hardship that every chemotherapy patient currently has to go through another frustrating thing about cancer is treatment that works for one person may not work for another but now Den noo protein design could change the game the NOA proteins are synthetic proteins designed with abilities not normally found in natural proteins they're easy to manufacture have new functions and are tunable to specific cases this means that we could create custom drugs or strategies for different kinds of patients possibly creating much higher success rates for treatments of serious diseases like cancer and finally let's zoom out of the human body and travel to the ocean I'm sure you can guess what that patch is that's the collection of Plastics in the ocean an estimated 50 to 75 trillion pieces of microplastics are in the ocean today microplastics in humans have already been found to heighten stress cause weight gain increased risk of cancer and disrupt reproductive development but now because of protein design we could have a solution and it all starts with this animal a strawberry anemon scientists took a defense protein called fragos toxin C or frag C from within this anemon it's a porous protein used to filter toxins meaning it's a perfect base for a protein capable of breaking down plastic let me show you why see if if you take frag C and attach just three amino acids you get an artificial protein that can degrade microplastics and nanoplastics allowing them to be broken down or recycled and since they're porous you could place these proteins in existing desalination plants this would allow these proteins to automatically eliminate the plastic particles that are currently making their way into some of our water supply and all of this was only possible with the Innovation provided by Alpha fult it gave us the ability to predict and model things like the enemity based proteins the plastic eating amino acids and the combination of the two now with a little bit more time we could have a viable solution to the microplastics problem responsible for some of the decline of our health and these exciting advancements are really just the beginning this technology has the potential to revolutionize how scientists approach biology and Medicine there are endless possibilities that this technology could help us reach and in this video we've really only scratched the surface so keep an eye out for the next revolutionary protein advancement it's coming sooner than you think

Channel: Chasing Down Chano

Views: 6,477

Keywords: alphafold, alphafold2, alphafold3, what is alphafold, google deepmind, deepmind, google ai, ai, artificial intelligence, machine learning, ted talk

Id: NB6MgJt-46A

Length: 13min 5sec (785 seconds)

Published: Tue Apr 23 2024