Claude read all 3,096 ASHG abstracts to surface 33 human genetics highlights
anshulkundaje · x · 2026-11-05
GWAS Signals author Veera M. Rajagopal used Claude to read all 3,096 abstracts from ASHG 2026 in Montreal, auto-flagging 189 matching the newsletter's focus areas (rare variants, GWAS, drug development, noncoding variants), then hand-curated 33 highlights across six themes: new drug targets, biobank-scale studies, founder populations, DNA repeats, directly tested noncoding variants, and genetic medicines.
Example finding: in 13,725 Helix exome-linked records, carriers of rare damaging GLP1R variants lost far less weight on semaglutide (2.4% vs 8.9% in matched controls), but tirzepatide still worked (13.3% vs 15.2%), implicating GIP receptor agonism as compensation.
The piece demonstrates an "AI first-pass + human curation" workflow for digesting conference-scale scientific literature.
More from Research
- Researchers praise Together Compiler for open discussion of self-improving infra tradeoffs — simran_s_arora · 2026-11-05
- Economist podcast: 'AI slop can swamp the scientific community,' peer review needs an overhaul — james_y_zou · 2026-11-05
- Thread wraps up: task interaction patterns persist in nonlinear networks on Multi-MNIST — ClementineDomi6 · 2026-11-05
- TMLR accepts paper analytically explaining task help vs interference in multi-task learning — ClementineDomi6 · 2026-11-05
- Fitted 0.8B model beats hosted 105B Sarvam Prism on custom decision taxonomy, 2.6x faster — HowDevelop · 2026-11-05
- Paper title hides method name ipSAE in Latin pun on 'the facts speak for themselves' — AllThingsApx · 2026-11-05