Twitxy Saved

Paul Ginart, MD PhD

Verified account This account is verified. Learn more

@paul_ginart

Building AI-Native Clinical Practice | Bioengineering, Longevity, Precision Medicine l EM Physician @MassGenBrigham
Boston, MA precisionlongevity.io Joined June 2015
422 Following 524 Followers 398 Posts

Latest posts 5

Paul Ginart, MD PhD Verified account @paul_ginart
Very cool model. I just ran the or so 4000 ultra rare variants in my own genome through this.

Luckily, nothing scary though I have a splice variant and a few IL-13 pathway hits which maybe explain my mild propensity toward eczema.

I鈥檓 sure there鈥檚 a lot more signal in rare but not well understood genetic diseases.
Google DeepMind @GoogleDeepMind Nov 11
Your heart, brain, and immune system share the same DNA sequence. So how can a single-letter change affect disease in some cells, while sparing the rest?

AlphaGenome 2 is our new AI model that zooms in on specific cells to help scientists understand what drives disease. 馃У
View quoted post on X
Paul Ginart, MD PhD Verified account @paul_ginart
But what does the data show about continuous glucose monitors and preventative metabolic health?

@michaelloesche and I are excited to share PLC's first research study, with pre-print live on medRxv.

We analyzed data from @Levels platform, looking at over 62,000 people with linked food logs, CGM traces, labs, step counts and more to see what CGM features are associated with metabolic health

1) CGM, via average glucose, outperforms HgbA1C at detecting elevated fasting insulin, particularly in younger healthy individuals, even after controlling for BMI

2) Glucose spike width but not spike height or frequency or carbohydrate sensitivity, carries independent metabolic signal. After controlling for average glucose, spike width retained an association with fasting insulin, triglycerides, and hs-CRP that the other spike features did not.

There are lots of plausible theories about what matters in longevity and prevention, and at PLC we dig into the data to distinguish the signal from the noise.

Pre-print link below:
13 replies 4 reposts 40 likes 8.1K views View on X
Paul Ginart, MD PhD Verified account @paul_ginart
People don鈥檛 realize it yet, but medical AI model interpretability is going to be THE rate limiting bottle neck a year from now, and last for at least 5 years even in a RSI scenario. Can鈥檛 under-write the liability of a prediction you can鈥檛 understand. Never a better time for physicians to train their own models