Cancer and antimicrobial resistance have the same solution

I'm typing this up at 10pm the day before I'm coming back to Cambridge, specifically because I want to escape the feeling that I'm not working on this website. It's just a first draft, but I just want to put something up on here now and I'll polish it later :)

Before my imagination was quenched by being trained to thinking "scientifically" (which comes with rigidity), I spent some time during my undergrad thinking about truly unique ways of solving some of the big problems in biology. Certainly the most recognisable one being that of cancer.

And throughout the teachings of the current understanding of cancer, I kept thinking, "this is all in the wrong direction!". It's not like I will exactly be first one to recognise what I'm about to explain to you, or to be the one to change the field. But as I progressed through my degree, it turns out that the ideas I will present are not unique, rather the field just does not have the easiest way of acting on it.

So what does the field do currently? As we progress through a biology degree, we're pushed through learning how three-letter-codes (e.g. "BCL-2") apparently "interact" with each other, represented by drawing a graphically-unpleasing map of these codes linked by a web of arrows, and how cancer is able to apparently break these somehow to take advantage of the cell's machinery. We speak about each of these components as if we know them inside-out, but the truth is, unless you have directly worked with them, we can't talk about each of these things much more deeply than, "oh right, that's part of the ERK pathway". We can't visualise what any of them really look like (unless you look up the 3D structure yourself), we can't predict exactly what will happen to any of them if you change something in the cell (unless you find someone having made a good mathematical model about a specific one, under specific conditions), and most importantly, you could not justify why each of these components are strictly necessary and why the cell did not opt for using something slightly different.

Further to the last point, as a biology degree progresses, there is a real point made about not anthropomorphising the 'work of evolution'. Evolution is the most powerful algorithm in the world, but it is just a simple result of having a replicator that subtly changes with every generation, along with some sort of 'selection pressure' (e.g. competition). None of its results are chosen by some sentient being.

We've been outcompeted, quite simply, in the world of antimicrobials because evolution will always find a biological solution against any problem it faces. What do we retaliate with? Our most successful drug-finding escapade thus far has been to roll the die, extremely efficiently. High-throughput-screening (HTS) is a general name for techniques with the goal to try millions of random molecules, where a few are expected to work a little, which can be tested further. Our 'search algorithm' was extremely successful for creating chemotherapeutic drugs, and we have made huge progress in some types of cancer, and we've found (and designed) various antibiotic types that have saved hundreds of millions of lives.

Yet, we're facing the reality of being outpaced. We're sounding the alarm of antimicrobial resistance, and how by 2050 there will be millions of deaths annually directly from resistant bacteria that would be treatable today. It turns out that, testing millions of molecules does not compete with bacteria's strategy of trying quadrillions of mutations every year, in the hopes of finding one that is resistant. (Further, evolution can be self-improving, whereas our methods are pretty static).

What people (i.e. the "public") doesn't seem to understand, is the same problem presents itself in cancer. There are a plethora of extremely treatable cancer types, yet most people are personally affected by cancer, whether themselves or someone close to them. What people don't understand further, is that the way we treat cancer currently is risky and quite literally trains cancer against our medicine.

Firstly, recognise that cancer is a disease of evolution. Most mutations are harmful, but rarely, they can be beneficial. For a cancer, this means only mutations that are beneficial for its own survival will survive, and the 'attempts' at mutations that fail die out.

Secondly, recognise this means each of the cells in the cancer will all be different from each other. We like to classify cancers from where they come from or specific characteristics, but the truth is, they're all unique, and each cell in each cancer is unique too. In the centre of a cancer, where there is less access to blood flow (as it's squished between the other cancer cells!), you'd expect mutations that help with the lack of oxygen or nutrients. Lack of oxygen also speeds up rates of mutation, which will make the central cells even more different!

Thirdly, recognise that this means cancer will evolve resistance against your own anti-cancer drugs. You treat a cancer, and it comes back in five years with a new resistance to the drug you used previously. Damn!

The truth is, our current pharmacological discovery process is on average safe, and on average effective, at the cost of being slow. It takes over ten years for a drug to be discovered, developed and tested, and finally released. How will this cure the cancers that will kill you in six months, that will evolve resistance to drugs you give it in a few weeks? How will this overcome antimicrobial resistance, when we haven't found a new class of antibiotics in 40 years?

Fixing this is difficult. How do you find, and test, these drugs extremely quickly and know for sure that they are effective and safe for the individual you are working with? That's why the field has been keeping on researching 'basic biological understanding', which involves the webs of proteins that are then taught to undergrads.

How do you beat evolution?

You could try to reverse evolution, and somehow use it yourself to find solutions to your problems. That is the... most straightforward, but nonetheless difficult option right now. I did email a professor about this idea, even saying, "am I crazy?", (has this not been thought of before?), and he did check my LinkedIn to see who I was, before choosing to not respond to my email. (I was just a silly first-year).

I did, however, contact a researcher from Sweden who had published a paper that described a 'skeleton protocol' for using bacterial evolution yourself to find new antibiotics. We hopped on a call, and the conclusion was that the idea could work, but the problem is one of scale. If you want to isolate each bacteria and grow them into further generations, but you swab each colony, that means you are effectively swabbing (maximum) a hundred possible bacteria per petri dish... of which a new mutation occurs every ~600 bacteria (figure from Claude). How do you get this to compete with the quadrillions all evolving at all times around the world against your antimicrobials? It's an impossibly tall task.

This could work long-term. We do have fields that has this problem: you have an input that changes a little, and the output can change a lot, and you're trying to get to a specific output, but you don't know what it is. These fields are encryption and (related) bitcoin mining! You cannot know what the solution is without attempting all the possibilities, of which there are an extraordinarily large amount. You must try them all, which is exactly what these quadrillions of bacteria are doing constantly, worldwide, at all times. The way we've solved it with bitcoin mining is firstly, optimisations in the compute (with the advent of ASICs, specialised computers just for mining bitcoin), and secondly, benefitting from improvements in tech in general to make these chips far far better, and thirdly, scaling up to have far more of these miners. I wonder if there's something to be learnt about how to find cancer or anti-bacterial therapeutics by looking at the lessons from these fields.

Certainly the closest thing we have is the folding@home projects, where users at home could donate their compute power and electricity to calculate whether a particular protein would indeed have the shape desired for whatever function. Nowadays, people would be looking into leveraging AI.

Guess what? There's this project called Evo, the latest of which being Evo 2, which is exactly this. A machine-learning-powered tool that can predict the various effects of small evolution steps. Should this be very accurate, we could, maybe, counter evolution with raw compute power?

Something like that.