Just published a new blog post!
If youโre interested in the latest research in medical AI and disease genetics give it a read ๐๏ธ
open.substack.com/pub/healthda...
Posts by Kira Detrois
New preprint ๐
Does diagnostic misclassification in EHR-based psychiatry GWAS suppress genetic discovery?
We use simulations and FinnGen data to expand the SuperControl framework and propose a new ML approach (PRISMA) to boost signal discovery.
Led by Lisa Eick & Zhiyu Yang!
๐งฌ๐ฅ Do the genetics that make you develop a disease also help you survive it? Not much.
Our new study in Nature Genetics including 9 disease and 7 biobanks shows:
โข Susceptibility variants โ survival
โข PRSs for onset weak at predicting progression
โข Lifespan PRS predicts survival better
A groundbreaking paper published @nature.com todayโthe first large health language modelโpredicts >1,000 diseases for a person and when.
Implications here:
erictopol.substack.com/p/dawn-of-a-...
Are you interested in pushing the limits of what population health "foundation model" could be used for? Would you like to study privacy and fairness of generative models that are trained on sensitive health data? Or combine this with your own ideas?
Two weeks left to apply to this PhD position ๐
Great review! I think itโs a good length.
I still learned to code in C++ and love it, but admittedly I also became fluent in R tidyverse in recent years, so that helps. But ever since I started using polars and parquet files I canโt stop talking about how amazing it is. As evidenced by this comment. Itโs just so logical to use and fast.
I am hiring! ๐
We offer a fully funded PhD student position to work on exciting machine learning projects utilizing rich, multimodal real-life health data, based at @fimm-uh.bsky.social in Helsinki, Finland, the happiest country in the world ๐ซ๐ฎ
jobs.helsinki.fi/job/Helsinki...
Feel free to DM me!
New @natgenet.nature.com study led by @tuomohartonen.bsky.social & @detroiki.bsky.social from FIMM explores how to better predict disease risk by combining information from electronic health records (phenotype risk scores) and genetics using 3 large biobank-based studies. Congratulations!
Excited to see my first paper out in Nature Genetics! ๐ We compared EHR-based models and polygenic scores for disease prediction across >845K participants in 3 biobank-based studies, as part of the INTERVENE consortium. Grateful to all collaborators who made this possible!