My take on AlphaFold2

I haven't spent too much time thinking about AlphaFold2 (AF2) since it entered the structural biology zeitgeist, but I was watching Veritasium's recent video on the topic and thought I would learn a bit more about it. For those who haven't heard of AlphaFold2, I highly recommend watching the video linked above, which explains AF2 better than I ever could. But the tl;dr version is this: AlphaFold2 is an artificial intelligence that takes any amino acid sequence (the building blocks of all proteins) as its input, and outputs the 3D protein structure it thinks that sequence is most likely to take. For some sequences, AlphaFold2 can do quite well at this notoriously complex task. On its debut in 2021, it was able to predict the 3D position of atoms in some protein backbones to within a few hundred nanometers!

There are a plethora of articles out there speculating about how AF2 is going to change biology and the world at large. But someone recently asked me my opinion: "How is AF2 going to change the world?" "What's the big deal?" 

The biggest impact I foresee for AF2 is that it will help scientists generate better ideas faster. To illustrate my point, let's pretend I am a geneticist who has discovered that a particular genetic mutation in humans causes some terrible syndrome, like complete short term memory loss. I would really like to design a drug that anyone with the mutation could take to cure them of this ailment. What steps might I take to begin tackling this problem, if I don't have access to AlphaFold2?

1. Identify the human protein that the mutation impacts and what that protein's general structure and function are. (Yes we're assuming the mutation is in a coding region and that we have some reasonable starting point for hypothesizing about its function.) Use those findings to design a well-informed experimental plan for screening a bunch of potential drug molecules for action against the mutant protein. Questions I might want to answer in this stage: Will I do an in vitro or in situ screening assay, or both? Will I screen small molecules? Peptides? Aptamers? Biologics? Am I looking for gain-of-function or loss-of-function results? Etc.
2. Probably pay to have the mutant DNA sequence synthesized, then genetically modify some bacteria with it so they produce the mutant protein for me.
3. Depending on findings from item 2, probably purify a large quantity of the mutant protein and use it for in vitro binding assays. Hopefully identify some molecules that bind to the protein.
4. Depending on findings from item 3, maybe go back to step 2 and start over. But if I'm lucky, use my purified protein for an in vitro functional assay. Hopefully develop a method to detect the protein's function, then test the molecules from Step 3 to see how they impact the protein's function.
5. Depending on findings from item 4, maybe go back to step 1 and start over. Otherwise, if things are still going well, get my protein of interest transfected into some human cells in a petri dish and develop an in situ functional assay. Hopefully the molecules identified in step 3 still work on the live cells I'm growing in the petri dish.
6. Depending on findings from item 5, maybe go back to step 1 and start over. Or, if I'm lucky, start thinking about designing an animal model that has a mutation analogous to the human one. Let's just assume it's going to be mice.
7. Spend either years or lots of money to generate a strain of mutant mice that expresses the mutant protein in all the same locations its observed in the human brain. Put those mice through behavioral tests to prove that the short-term memory syndrome is well-modeled by the mice. Hopefully I don't have to side-quest here to design some new mouse behavioral assays. 
8. Depending on findings from item 7, probably go back to step 1 and start over (mice aren't humans after all, and how do you test a mouse's short term memory anyway??). If I'm still getting lucky at this stage, I'd probably design some blinded, controlled studies where I determine if the molecule(s) from Step 3 have any effect on the mutant mice with short term memory loss, and if they cause any unwanted side effects or toxicity. Fingers crossed.
9. Depending on findings from item 8, almost certainly go back to step 1 and start over. Or give up on curing this disease because I'm nearing retirement. Or since I've come all this way and seem to have identified a molecule that does something promising in mice, design a Phase 1 clinical trial to see if the drug is safe for humans to take. If so, move on to step 10!
10. Design Phase 2 clinical trials to see if the molecule is effective in treating patients who actually have the memory loss mutation. Continue monitoring for side effects and toxicity of course.
11. Design Phase 3 clinical trials to see if the drug is still safe and efficacious when given to a larger and more diverse group of patients. I hope I started the FDA-submission paperwork back in step 8...
12. Submit the application for a new drug candidate and wait. Go back to whatever step the FDA says needs more data....or if I'm lucky, my molecule is approved and can be used to cure people with the memory-wiping mutation! Huzzah!

So then, how does the process above change if I do have access to AlphaFold2? 

It doesn't. 

In fact, the only step where AlphaFold2 has a major impact in the process above is in Step 1. But Step 1 is the most important step. It's where I use any and every tool I have available to me to understand as much as I can about the mutant protein that causes the syndrome. I rely on all of that knowledge to plan experiments that are most likely to lead to a successful drug candidate, and an eventual cure. Any edge I gain in Step 1 has the potential to save me (and patients) years of time and lots of money by increasing the likelihood of success in Steps 2-12. And that's why people say AlphaFold2 can change the world. Because when it works, it gives me information in Step 1 'for free' that could change the course of the next decade of my work. 

So how is AlphaFold2 going to change the world first? My answer is that I don't know. No one knows which "Step 1s" will lead to Earth-shattering breakthroughs - like Alzheimer's cures, or carbon sequestration, or biodegradable plastic. But I can guarantee that AF2 will not be solving problems like world hunger on its own. Behind any AF2-supported breakthroughs there will be at least a small army of human scientists verifying that anything AF2 generates is consistent with reality. They will do this with technically challenging and expensive experimentation and with dedication in the face of many setbacks. And that's what will change the world.