#258 Artificial Intelligence for Genetic Testing and DNA Analysis - Joe Cohen

Joe Cohen (0:00)

All of my major investments have been in health, just in improving my health. That in turn helps me to be happier, right? Like I'm just in a better mood. Having a blueprint and testing is a very good return on investment.

Siim Land (0:15)

Welcome to the Siim Land Podcast. I'm your host Siim Land, and today our guest is Joe Cohen. Joe is the founder and CEO of selfhacked.com and SelfDecode. SelfDecode is a genetics decoding company. You can get personalized health recommendations based on your DNA and the latest scientific research. They have numerous different DNA reports for different areas of focus like weight loss, longevity, gastrointestinal health, cognition and even mood. Recently SelfDecode came out with their 2 software that incorporates artificial intelligence in generating the DNA reports.

It's the most advanced and comprehensive consumer DNA service in the world. You can get a 10% discount with the code Siim at get.selfdecode.com forward slash siim.

Joe, welcome to the show.

Joe Cohen (0:58)

Hey, Siim, thanks for having me.

Siim Land (1:00)

Yeah, I'm glad to talk with you again, and we spoke first time last year in around maybe like the spring.

So it's been like a year since the last time we talked about DNA. So has there been like any new discoveries about this DNA tests and genetics? Or how is it like the technology itself? Has it evolved any further?

Joe Cohen (1:21)

Yeah, so, you know, I'm obviously a big fan of using DNA and biohacking and learning about, you know, your body, right? So biohacking is all about learning and optimizing your body. And I think, you know, using personalized health is very big for optimizing, learning and optimizing your body. So there's some exciting updates in the DNA field, and I feel like that's something that we could talk about a bit. But basically, you know, up to now, companies, you know, companies have been using a handful of variants or SNPs to basically look at someone's risk score, right?

And, well, I mean, for the past two years, we've been working on basically the next generation analysis. So let me just explain to you what, you know, so basically there's kind of like there's a level of statistical significance, and what other companies are doing is they're looking at maybe the most statistically significant variants or SNPs, and they're putting it into some algorithm. Every company has a different quality algorithm. They might have different quality SNPs, but more or less they're looking at a study, they're looking at the most significant variants or multiple studies, and then they're putting it through some kind of algorithm. That's more or less what is done. Now, basically in 2018, there was a famous Harvard paper that got published about, I mean, there were papers published also in Stanford that basically just using a handful of SNPs is not very effective.

And the reason is because it turns out that millions of variants are having an impact on every trait, right? Most traits, when it comes to chronic diseases and maybe not eye color or something like that or hair color, those are more controlled by much fewer variants. But when it comes to something that's very complex, such as intelligence or cognitive function or your cardiovascular system, your heart health, any kind of complex trait or chronic disease, the chronic diseases are usually complex traits, they're controlled by millions of variants usually, or hundreds of thousands. And so the next generation approach is to be looking at hundreds of thousands or millions of variants and using machine learning and AI to look at the, basically find the trend in individual or thousands of individuals with this, with these variants, these millions of variants have one thing, and thousands of individuals with millions other variants or whatever. And it basically looks for trends and see where, what are, it's not just having one variant, it's having a trend of variants and then finding what the risk goers are based on those trends. And so basically, you know, so when, if you're looking into a DNA testing company, you should be looking at how many variants they're looking at. And I looked at every single one and they're only looking, I was actually pretty surprised, because most of them tell you, if they don't tell you, then they're not actually looking at many.

But almost all of them show, the ones that don't show are usually even worse, but it basically goes from anywhere between 1 and 30 variants, right? Or something like that. It's a low number of variants, anywhere less than 100

And so what we're doing now is we're actually looking, we put on every report how many variants we looked at, and so some of them are above a million. They're usually in the hundreds of thousands or above a million. And that's a very compute-heavy process. So it was very difficult to do this. You need data engineering teams, and you need very big data engineering teams, software development teams, and also data scientists and top scientists in order to pull it off. And so that's one big thing that I think is very important to understand about the field of DNA now. And it's something I only realized relatively recently, that you really do have to look at a large number of these variants.