[after_care]

> Maybe in the future AI will solve this problem, if we can learn to accurately model the human body. But given it's insane complexity that's certainly going to be a big challenge

It’s clear to me that modeling the human body would be complicated, but I don’t have any idea about the order of magnitude. Can someone chime in?

[tsimionescu]

Well, to model organic interactions you basically need a molecule-level model of the body or even deeper - especially for extremely complex molecules like proteins (which can famously "fold" in extremely hard to predict ways).

So, to very accurately model the human body, you need a model that has as many parameters as there are molecules in a human body, give or take a few orders of magnitude.

I have no idea to what extent you could get away with more high-level models - i would expect they could work well for some kinds of problems, but not for very systemic interactions, like hormonal or aging-related issues, that affect each cell in the human body to some extent.

[staticassertion]

I really doubt you'd start with "Let's emulate every atom of the human body". That's a huge waste of time. The atoms in my toenail are not relevant to most models you'd care about.

You'd probably start off instead by emulating specific proteins. That's already really hard, but obviously a trillion times easier than what you're suggesting.

Once you can emulate a number of proteins you can start to emulate interactions of drugs with those proteins. That's probably going to get you to a good enough heuristic where you can say "ok let's try this on something alive" or "this clearly doesn't work".

This is, to my knowledge, the existing approach being taken. It's just that a single protein is extremely complex. A single protein is a series of amino acids linked by peptides - each amino acid is itself a pairing of numerous atoms.

As you go from atom -> amino acid -> poly peptide -> protein you get a massive increase in complexity.

[tsimionescu]

I thought the question was about developing new drugs entirely using computer models, at least until the first human trials. If that is the goal, we would need a model of the proteins in the entire body to get an idea of how a substance could affect every protein they interact with in every organ they reach, what organs it could actually reach, and how the modifications to each protein would in turn cascade.

And the toe nails and similar things are why I was saying "give or take a few orders of magnitude". Not to mention, certain drugs could affect your toenail.

[staticassertion]

I still think it would be unnecessary to do anything more than what I described, but just at a larger scale. But not the scale of emulating a whole human being. For the most part you're going to be looking for "does this bind with X but not Y" and expanding your set of X and Y is what makes things safer.

I think you could get to human trials using just a computer if you had a sufficiently sized set of emulated X's and Y's.

[tsimionescu]

Well, if it binds with X, what prevents X+substance from binding with Y or Z?

[staticassertion]

Yeah you might want to add a few levels there for sure. But that's still a far cry from fully emulating the human body.

[rmbyrro]

I guess the most difficult part is modeling all possible interactions related to an infinity of chemicals, DNA variations, environment influences, past medical issues, etc, etc.

I bet this would require computational power equivalent to whatever is "processing" the govern of real-world interactions in our bodies.