Excited to share our new paper in @JNCI_Now! We integrated GWAS data from 11 solid cancers with ~1,500 cell type annotations to pinpoint WHERE in the body cancer risk variants actually act. A thread 👇
@peter-kraft.bsky.social
📣 𝐖𝐞’𝐫𝐞 𝐡𝐢𝐫𝐢𝐧𝐠 𝐚 𝐩𝐨𝐬𝐭𝐝𝐨𝐜 𝐢𝐧 𝐩𝐨𝐩𝐮𝐥𝐚𝐭𝐢𝐨𝐧 & 𝐬𝐭𝐚𝐭𝐢𝐬𝐭𝐢𝐜𝐚𝐥 𝐠𝐞𝐧𝐞𝐭𝐢𝐜𝐬! 🌍📈🧬
If you’re finishing a PhD (or know someone who is) and want to work on complex trait biology + 𝑟𝑒𝑎𝑙-𝑤𝑜𝑟𝑙𝑑 𝑖𝑚𝑝𝑎𝑐𝑡, read on 👇
Introducing the 𝐎𝐩𝐞𝐧 𝐒𝐜𝐡𝐨𝐥𝐚𝐫𝐬𝐡𝐢𝐩 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐟𝐨𝐫 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡𝐞𝐫𝐬 𝐒𝐞𝐫𝐢𝐞𝐬
We’re starting with 2 online courses, with more on the way:
→ Fundamentals of Open Scholarship > bit.ly/4vYZVjh
→ Preregistration and Reregistration Reports > bit.ly/48sPpXE
🧵 (1/6)
🎉New preprint: Phenome-derived polygenic scores and social determinants jointly shape context-dependent disease risk.
We evaluate disease risk along 2 complementary axes:
• how genetic liability is represented
• the social context in which it is expressed
“[G]enerative models generate numbers which are thought to represent real possibilites in the world. It is very tempting to change model inputs and then look at the result. A safe way to describe this is ‘in silico perturbation’, but many use the word ‘counterfactual.’ This is very dangerous.”
(Quotes lightly edited for space. Any errors on me. Do read the original post.)
This👇 A key to any productive collaboration between computer scientists and epidemiologists (or clinicians or biologists or…) is to learn each others’ languages. Sometimes we use different words for the same concepts, sometimes we use the same words but mean very different things.
(Even ‘in silico perturbation’ might be too strong. Maybe simply ‘simulation’? Generative models capture more complexity [tuned to training data] than your typical biostats simulation [‘now we’re really going to get fancy and throw in 2nd order interactions’], but they’re still simulacra.)
Kodama: genotype compression and matrix multiplication leveraging genetic relatedness www.biorxiv.org/content/10.6...
“Many valid, tested, robust, and clearly generalisable generative models will not correspond to the real world if you do this in silico perturbation. The presence of a robust and generalisable association model absolutely does not mean that interventions can be modeled.”