Looking for some feedback to pressure test an idea: Accelerating the creation of an AI model of human aging?

#1

Would love some feedback on whether people think something like this has value.

I know there are already a few longitudinal studies, but want to know if some specific to aging might be useful.

I’ve been thinking about the bottlenecks in longevity research and I’ve been wondering if there a way to accelerated it with AI.

It looks as if we don’t have anywhere near enough longitudinal human data.

My idea is to recruit cohorts of adults at five-year intervals for five years from say, 20, onwards all the way through to 80+ and follow all of them simultaneously.

You’d to collect as much data as possible. Proteomics, metabolomics, immune profiling, epigenetics, imaging, DEXA, cognitive and functional testing, continuous wearable data, microbiome, etc. Basically, everything that is available.

Then you’d have overlapping longitudinal trajectories across most of adult lifespan.

Instead of having five years of data from one age group, you’d have overlapping data spanning roughly 20–85.

I know this isn’t equivalent to following the same people for 60 years. But this would at least be a start and would be fairly quick, plus, if the cohorts roll over, then it starts to become much more valuable over time.

My question, as a non-scientist who spends far too much time thinking about things I can’t do anything about, is whether something designed specifically like this could become training data for AI models of human aging?

I know there’s the UK Biobank, but I’m imagining something deliberately designed from day one designed to track aging trajectories.

I have no idea how to do this. I’m throwing the idea out specifically to see if anyone thinks this is a good idea .

Would this scientifically useful? Is there already a project doing essentially this? What are the flaws? How much and how complicated would something like this be?

And if anyone were to design a dataset for a foundation model of human aging, what would need to be in it?

1 Like
#2

These are all good ideas and their essence is being explored in various ways. The constraints researchers face when contemplating such studies are many and varied. The largest constraint is financial. Projects of this nature are very expensive to establish, operate, and manage. A close second is human factors affecting the establishment and management of cohorts. The metrics you would like to gather are intrusive and the typical person would not be interested, This means you will be recruiting atypical participants or paying “typical” participants. Either way, biases are created which must be controlled for to the degree possible or as much as the project can afford. Because large numbers of participants will drop out (or die) over time, the initial cohorts will have to be very large to ensure significance. Large, long term studies must address many other challenges but these first two are often show stoppers and why most scientists look for clever ways to approximate components of an ideal study using extant data or opportunistic piggybacking on other projects.

As for your basic design wireframe, if designed and executed well, it would – with some obvious exceptions such as inconveniently timed pandemics – approximate the equivalence of studying a single cohort for 60 years.

1 Like
#3

I am always happy if there are other people who want to join me in my biohacking, but it involves quite a bit of testing and can be a bit edgy.

2 Likes