How does population level targeted proteomics do in predicting cardiovascular events?

 

Unless you've been living under a rock where you don't hear much about proteomics, you probably know that a bunch of people did plasma proteomics on 53,000 samples with proximity extension assays (O-link panels, not the little 96 target ones, the big ones that require super high end RNA Sequencers). 

Here is one of the original papers (October 2023). There have been some really positive statements about how the O-link data correlates well in some regards with GWAS level findings. 

Now this data is basically out there for people who want to explore it. You can find it here, but keep this in mind.

Excluding GWAS associations, how does it currently look in actual practice? There are going to be tons of thse out here, but this one is really fun. 

Realistically, I like this because it's short and it is focused on current clinical markers of cardiac stress, which were old when I was running colorimetric assays in the clinics from 2003-2007. They're still in use because they're pretty darned good. 

This group took the O-link data and did some very fancy sounding machine learning stuff I couldn't possibly explain or replicate and had the algorithm predict if someone would havea major cardiac event. 

And this is what is SO COOL about a dataset this size. They actually have samples from people over time. They can separate the same patients that they have proteomics on in the past and the future. Then you can see if the smart machine was right or wrong. And you have the original clinical data from regular old comprehensive metabolic or cardiac specific panels for both. 

How'd it do? The authors words are better than mine. 

Don't get me wrong, this resource is awesome. And maybe a different machine learning tool would do better. And if our cardiac markers in the clinic weren't as good at they are (for identifying cardiac events in men, they aren't nearly as adequate in women, but that's a completely different topic for more qualified people), they wouldn't still be in use. 

It is, however, interesting that when we apply these data to something concrete, it isn't quite the world shaking advance that we might have hoped from the genome wide association studies.