Excited to share our recent paper in #npjBreastCancer! We investigated what drives the AI-generated mammogram risk score (MRS)—a texture feature that predicts breast cancer risk beyond breast density. A thread 👇 @peter-kraft.bsky.social
MRS strongly predicts breast cancer in 853 NHS II women: OR=1.92/SD, 10-yr AUC=0.69. After adjusting for BI-RADS density, MRS stayed strong (OR=1.85)—but BI-RADS effect was substantially attenuated after adjusting for MRS. MRS captures risk beyond density alone!
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
What's associated with MRS? Higher breast density & benign breast disease history ↑ MRS; higher BMI & early-life body size ↓ MRS. After adjusting for density, all three associations attenuated.
Mendelian randomization confirmed: genetics driving higher breast density causally ↑ MRS. Bonus: after adjusting for density, central obesity (WHRadjBMI) also significantly ↑ MRS. Full paper: doi.org/10.1038/s415...
Variant by variant, we built 489 regulatory quadruplets: causal SNP → cell context → target gene → cancer. Some genes like CDKN1A are implicated across cancers through distinct contexts. Code + data: github.com/xueyaowunci/...; github.com/ArtemKimUSC/...
Standard enrichment flagged 141 contexts. Our cell type fine mapping (CT-FM) framework distilled these to just 4 high confidence biological contexts: mammary luminal epithelial cells for breast, VCaP for prostate, COAD for colorectal, KIRC for renal.
